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/mm.h>
  30 #include <linux/swap.h>
  31 #include <linux/wait.h>
  32 #include <linux/mutex.h>
  33 #include <linux/device.h>
  34
  35 #include <lttng-events.h>
  36 #include <lttng-tracer.h>
  37 #include <wrapper/irqdesc.h>
  38 #include <wrapper/fdtable.h>
  39 #include <wrapper/namespace.h>
  40 #include <wrapper/irq.h>
  41 #include <wrapper/tracepoint.h>
  42 #include <wrapper/genhd.h>
  43 #include <wrapper/file.h>
  44 #include <wrapper/fdtable.h>
  45 #include <wrapper/sched.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 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_network_interface,
 138         TP_PROTO(struct lttng_session *session,
 139                 struct net_device *dev, struct in_ifaddr *ifa),
 140         TP_ARGS(session, dev, ifa));
 141
 142 #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
 143 LTTNG_DEFINE_TRACE(lttng_statedump_cpu_topology,
 144         TP_PROTO(struct lttng_session *session, struct cpuinfo_x86 *c),
 145         TP_ARGS(session, c));
 146 #endif
 147
 148 struct lttng_fd_ctx {
 149         char *page;
 150         struct lttng_session *session;
 151         struct files_struct *files;
 152 };
 153
 154 /*
 155  * Protected by the trace lock.
 156  */
 157 static struct delayed_work cpu_work[NR_CPUS];
 158 static DECLARE_WAIT_QUEUE_HEAD(statedump_wq);
 159 static atomic_t kernel_threads_to_run;
 160
 161 enum lttng_thread_type {
 162         LTTNG_USER_THREAD = 0,
 163         LTTNG_KERNEL_THREAD = 1,
 164 };
 165
 166 enum lttng_execution_mode {
 167         LTTNG_USER_MODE = 0,
 168         LTTNG_SYSCALL = 1,
 169         LTTNG_TRAP = 2,
 170         LTTNG_IRQ = 3,
 171         LTTNG_SOFTIRQ = 4,
 172         LTTNG_MODE_UNKNOWN = 5,
 173 };
 174
 175 enum lttng_execution_submode {
 176         LTTNG_NONE = 0,
 177         LTTNG_UNKNOWN = 1,
 178 };
 179
 180 enum lttng_process_status {
 181         LTTNG_UNNAMED = 0,
 182         LTTNG_WAIT_FORK = 1,
 183         LTTNG_WAIT_CPU = 2,
 184         LTTNG_EXIT = 3,
 185         LTTNG_ZOMBIE = 4,
 186         LTTNG_WAIT = 5,
 187         LTTNG_RUN = 6,
 188         LTTNG_DEAD = 7,
 189 };
 190
 191
 192 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,11,0))
 193
 194 #define LTTNG_PART_STRUCT_TYPE struct block_device
 195
 196 static
 197 int lttng_get_part_name(struct gendisk *disk, struct block_device *part, char *name_buf)
 198 {
 199         const char *p;
 200
 201         p = bdevname(part, name_buf);
 202         if (!p)
 203                 return -ENOSYS;
 204
 205         return 0;
 206 }
 207
 208 static
 209 dev_t lttng_get_part_devt(struct block_device *part)
 210 {
 211         return part->bd_dev;
 212 }
 213
 214 #else
 215
 216 #define LTTNG_PART_STRUCT_TYPE struct hd_struct
 217
 218 static
 219 int lttng_get_part_name(struct gendisk *disk, struct hd_struct *part, char *name_buf)
 220 {
 221         const char *p;
 222         struct block_device bdev;
 223
 224         /*
 225          * Create a partial 'struct blockdevice' to use
 226          * 'bdevname()' which is a simple wrapper over
 227          * 'disk_name()' but has the honor to be EXPORT_SYMBOL.
 228          */
 229         bdev.bd_disk = disk;
 230         bdev.bd_part = part;
 231
 232         p = bdevname(&bdev, name_buf);
 233         if (!p)
 234                 return -ENOSYS;
 235
 236         return 0;
 237 }
 238
 239 static
 240 dev_t lttng_get_part_devt(struct hd_struct *part)
 241 {
 242         return part_devt(part);
 243 }
 244 #endif
 245
 246 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,12,0))
 247 static
 248 int lttng_statedump_each_block_device(struct lttng_session *session, struct gendisk *disk)
 249 {
 250         struct block_device *part;
 251         unsigned long idx;
 252         int ret = 0;
 253
 254         /* Include partition 0 */
 255         idx = 0;
 256
 257         rcu_read_lock();
 258         xa_for_each(&disk->part_tbl, idx, part) {
 259                 char name_buf[BDEVNAME_SIZE];
 260
 261                 /* Exclude non-partitions bdev and empty partitions. */
 262                 if (bdev_is_partition(part) && !bdev_nr_sectors(part))
 263                         continue;
 264
 265                 if (lttng_get_part_name(disk, part, name_buf) == -ENOSYS) {
 266                         ret = -ENOSYS;
 267                         goto end;
 268                 }
 269                 trace_lttng_statedump_block_device(session, lttng_get_part_devt(part),
 270                                 name_buf);
 271         }
 272 end:
 273         rcu_read_unlock();
 274         return ret;
 275 }
 276 #else
 277 static
 278 int lttng_statedump_each_block_device(struct lttng_session *session, struct gendisk *disk)
 279 {
 280         struct disk_part_iter piter;
 281         LTTNG_PART_STRUCT_TYPE *part;
 282
 283         disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
 284
 285         while ((part = disk_part_iter_next(&piter))) {
 286                 char name_buf[BDEVNAME_SIZE];
 287
 288                 if (lttng_get_part_name(disk, part, name_buf) == -ENOSYS) {
 289                         disk_part_iter_exit(&piter);
 290                         return -ENOSYS;
 291                 }
 292                 trace_lttng_statedump_block_device(session, lttng_get_part_devt(part),
 293                                 name_buf);
 294         }
 295         disk_part_iter_exit(&piter);
 296
 297         return 0;
 298 }
 299 #endif
 300
 301 static
 302 int lttng_enumerate_block_devices(struct lttng_session *session)
 303 {
 304         struct class *ptr_block_class;
 305         struct device_type *ptr_disk_type;
 306         struct class_dev_iter iter;
 307         struct device *dev;
 308         int ret = 0;
 309
 310         ptr_block_class = wrapper_get_block_class();
 311         if (!ptr_block_class) {
 312                 ret = -ENOSYS;
 313                 goto end;
 314         }
 315         ptr_disk_type = wrapper_get_disk_type();
 316         if (!ptr_disk_type) {
 317                 ret = -ENOSYS;
 318                 goto end;
 319         }
 320         class_dev_iter_init(&iter, ptr_block_class, NULL, ptr_disk_type);
 321         while ((dev = class_dev_iter_next(&iter))) {
 322                 struct gendisk *disk = dev_to_disk(dev);
 323
 324                 /*
 325                  * Don't show empty devices or things that have been
 326                  * suppressed
 327                  */
 328                 if (get_capacity(disk) == 0 ||
 329                     (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
 330                         continue;
 331
 332                 ret = lttng_statedump_each_block_device(session, disk);
 333         }
 334         class_dev_iter_exit(&iter);
 335 end:
 336         return ret;
 337 }
 338
 339 #ifdef CONFIG_INET
 340
 341 static
 342 void lttng_enumerate_device(struct lttng_session *session,
 343                 struct net_device *dev)
 344 {
 345         struct in_device *in_dev;
 346         struct in_ifaddr *ifa;
 347
 348         if (dev->flags & IFF_UP) {
 349                 in_dev = in_dev_get(dev);
 350                 if (in_dev) {
 351                         for (ifa = in_dev->ifa_list; ifa != NULL;
 352                              ifa = ifa->ifa_next) {
 353                                 trace_lttng_statedump_network_interface(
 354                                         session, dev, ifa);
 355                         }
 356                         in_dev_put(in_dev);
 357                 }
 358         } else {
 359                 trace_lttng_statedump_network_interface(
 360                         session, dev, NULL);
 361         }
 362 }
 363
 364 static
 365 int lttng_enumerate_network_ip_interface(struct lttng_session *session)
 366 {
 367         struct net_device *dev;
 368
 369         read_lock(&dev_base_lock);
 370         for_each_netdev(&init_net, dev)
 371                 lttng_enumerate_device(session, dev);
 372         read_unlock(&dev_base_lock);
 373
 374         return 0;
 375 }
 376 #else /* CONFIG_INET */
 377 static inline
 378 int lttng_enumerate_network_ip_interface(struct lttng_session *session)
 379 {
 380         return 0;
 381 }
 382 #endif /* CONFIG_INET */
 383
 384 static
 385 int lttng_dump_one_fd(const void *p, struct file *file, unsigned int fd)
 386 {
 387         const struct lttng_fd_ctx *ctx = p;
 388         const char *s = d_path(&file->f_path, ctx->page, PAGE_SIZE);
 389         unsigned int flags = file->f_flags;
 390         struct fdtable *fdt;
 391
 392         /*
 393          * We don't expose kernel internal flags, only userspace-visible
 394          * flags.
 395          */
 396         flags &= ~FMODE_NONOTIFY;
 397         fdt = files_fdtable(ctx->files);
 398         /*
 399          * We need to check here again whether fd is within the fdt
 400          * max_fds range, because we might be seeing a different
 401          * files_fdtable() than iterate_fd(), assuming only RCU is
 402          * protecting the read. In reality, iterate_fd() holds
 403          * file_lock, which should ensure the fdt does not change while
 404          * the lock is taken, but we are not aware whether this is
 405          * guaranteed or not, so play safe.
 406          */
 407         if (fd < fdt->max_fds && lttng_close_on_exec(fd, fdt))
 408                 flags |= O_CLOEXEC;
 409         if (IS_ERR(s)) {
 410                 struct dentry *dentry = file->f_path.dentry;
 411
 412                 /* Make sure we give at least some info */
 413                 spin_lock(&dentry->d_lock);
 414                 trace_lttng_statedump_file_descriptor(ctx->session,
 415                         ctx->files, fd, dentry->d_name.name, flags,
 416                         file->f_mode);
 417                 spin_unlock(&dentry->d_lock);
 418                 goto end;
 419         }
 420         trace_lttng_statedump_file_descriptor(ctx->session,
 421                 ctx->files, fd, s, flags, file->f_mode);
 422 end:
 423         return 0;
 424 }
 425
 426 /* Called with task lock held. */
 427 static
 428 void lttng_enumerate_files(struct lttng_session *session,
 429                 struct files_struct *files,
 430                 char *tmp)
 431 {
 432         struct lttng_fd_ctx ctx = { .page = tmp, .session = session, .files = files, };
 433
 434         lttng_iterate_fd(files, 0, lttng_dump_one_fd, &ctx);
 435 }
 436
 437 #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
 438 static
 439 int lttng_enumerate_cpu_topology(struct lttng_session *session)
 440 {
 441         int cpu;
 442         const cpumask_t *cpumask = cpu_possible_mask;
 443
 444         for (cpu = cpumask_first(cpumask); cpu < nr_cpu_ids;
 445                         cpu = cpumask_next(cpu, cpumask)) {
 446                 trace_lttng_statedump_cpu_topology(session, &cpu_data(cpu));
 447         }
 448
 449         return 0;
 450 }
 451 #else
 452 static
 453 int lttng_enumerate_cpu_topology(struct lttng_session *session)
 454 {
 455         return 0;
 456 }
 457 #endif
 458
 459 #if 0
 460 /*
 461  * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
 462  * (scheduling in atomic). Normally, the tasklist lock protects this kind of
 463  * iteration, but it is not exported to modules.
 464  */
 465 static
 466 void lttng_enumerate_task_vm_maps(struct lttng_session *session,
 467                 struct task_struct *p)
 468 {
 469         struct mm_struct *mm;
 470         struct vm_area_struct *map;
 471         unsigned long ino;
 472
 473         /* get_task_mm does a task_lock... */
 474         mm = get_task_mm(p);
 475         if (!mm)
 476                 return;
 477
 478         map = mm->mmap;
 479         if (map) {
 480                 down_read(&mm->mmap_sem);
 481                 while (map) {
 482                         if (map->vm_file)
 483                                 ino = map->vm_file->lttng_f_dentry->d_inode->i_ino;
 484                         else
 485                                 ino = 0;
 486                         trace_lttng_statedump_vm_map(session, p, map, ino);
 487                         map = map->vm_next;
 488                 }
 489                 up_read(&mm->mmap_sem);
 490         }
 491         mmput(mm);
 492 }
 493
 494 static
 495 int lttng_enumerate_vm_maps(struct lttng_session *session)
 496 {
 497         struct task_struct *p;
 498
 499         rcu_read_lock();
 500         for_each_process(p)
 501                 lttng_enumerate_task_vm_maps(session, p);
 502         rcu_read_unlock();
 503         return 0;
 504 }
 505 #endif
 506
 507 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
 508
 509 static
 510 int lttng_list_interrupts(struct lttng_session *session)
 511 {
 512         unsigned int irq;
 513         unsigned long flags = 0;
 514         struct irq_desc *desc;
 515
 516 #define irq_to_desc     wrapper_irq_to_desc
 517         /* needs irq_desc */
 518         for_each_irq_desc(irq, desc) {
 519                 struct irqaction *action;
 520                 const char *irq_chip_name =
 521                         irq_desc_get_chip(desc)->name ? : "unnamed_irq_chip";
 522
 523                 local_irq_save(flags);
 524                 raw_spin_lock(&desc->lock);
 525                 for (action = desc->action; action; action = action->next) {
 526                         trace_lttng_statedump_interrupt(session,
 527                                 irq, irq_chip_name, action);
 528                 }
 529                 raw_spin_unlock(&desc->lock);
 530                 local_irq_restore(flags);
 531         }
 532         return 0;
 533 #undef irq_to_desc
 534 }
 535 #else
 536 static inline
 537 int lttng_list_interrupts(struct lttng_session *session)
 538 {
 539         return 0;
 540 }
 541 #endif
 542
 543 /*
 544  * Statedump the task's namespaces using the proc filesystem inode number as
 545  * the unique identifier. The user and pid ns are nested and will be dumped
 546  * recursively.
 547  *
 548  * Called with task lock held.
 549  */
 550 static
 551 void lttng_statedump_process_ns(struct lttng_session *session,
 552                 struct task_struct *p,
 553                 enum lttng_thread_type type,
 554                 enum lttng_execution_mode mode,
 555                 enum lttng_execution_submode submode,
 556                 enum lttng_process_status status)
 557 {
 558         struct nsproxy *proxy;
 559         struct pid_namespace *pid_ns;
 560         struct user_namespace *user_ns;
 561
 562         /*
 563          * The pid and user namespaces are special, they are nested and
 564          * accessed with specific functions instead of the nsproxy struct
 565          * like the other namespaces.
 566          */
 567         pid_ns = task_active_pid_ns(p);
 568         do {
 569                 trace_lttng_statedump_process_pid_ns(session, p, pid_ns);
 570                 pid_ns = pid_ns ? pid_ns->parent : NULL;
 571         } while (pid_ns);
 572
 573
 574         user_ns = task_cred_xxx(p, user_ns);
 575         do {
 576                 trace_lttng_statedump_process_user_ns(session, p, user_ns);
 577                 /*
 578                  * trace_lttng_statedump_process_user_ns() internally
 579                  * checks whether user_ns is NULL. While this does not
 580                  * appear to be a possible return value for
 581                  * task_cred_xxx(), err on the safe side and check
 582                  * for NULL here as well to be consistent with the
 583                  * paranoid behavior of
 584                  * trace_lttng_statedump_process_user_ns().
 585                  */
 586                 user_ns = user_ns ? user_ns->lttng_user_ns_parent : NULL;
 587         } while (user_ns);
 588
 589         /*
 590          * Back and forth on locking strategy within Linux upstream for nsproxy.
 591          * See Linux upstream commit 728dba3a39c66b3d8ac889ddbe38b5b1c264aec3
 592          * "namespaces: Use task_lock and not rcu to protect nsproxy"
 593          * for details.
 594          */
 595 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(3,17,0) || \
 596                 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
 597                 LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \
 598                 LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
 599         proxy = p->nsproxy;
 600 #else
 601         rcu_read_lock();
 602         proxy = task_nsproxy(p);
 603 #endif
 604         if (proxy) {
 605 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(4,6,0))
 606                 trace_lttng_statedump_process_cgroup_ns(session, p, proxy->cgroup_ns);
 607 #endif
 608                 trace_lttng_statedump_process_ipc_ns(session, p, proxy->ipc_ns);
 609 #ifndef LTTNG_MNT_NS_MISSING_HEADER
 610                 trace_lttng_statedump_process_mnt_ns(session, p, proxy->mnt_ns);
 611 #endif
 612                 trace_lttng_statedump_process_net_ns(session, p, proxy->net_ns);
 613                 trace_lttng_statedump_process_uts_ns(session, p, proxy->uts_ns);
 614         }
 615 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(3,17,0) || \
 616                 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
 617                 LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \
 618                 LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
 619         /* (nothing) */
 620 #else
 621         rcu_read_unlock();
 622 #endif
 623 }
 624
 625 static
 626 int lttng_enumerate_process_states(struct lttng_session *session)
 627 {
 628         struct task_struct *g, *p;
 629         char *tmp;
 630
 631         tmp = (char *) __get_free_page(GFP_KERNEL);
 632         if (!tmp)
 633                 return -ENOMEM;
 634
 635         rcu_read_lock();
 636         for_each_process(g) {
 637                 struct files_struct *prev_files = NULL;
 638
 639                 p = g;
 640                 do {
 641                         enum lttng_execution_mode mode =
 642                                 LTTNG_MODE_UNKNOWN;
 643                         enum lttng_execution_submode submode =
 644                                 LTTNG_UNKNOWN;
 645                         enum lttng_process_status status;
 646                         enum lttng_thread_type type;
 647                         struct files_struct *files;
 648
 649                         task_lock(p);
 650                         if (p->exit_state == EXIT_ZOMBIE)
 651                                 status = LTTNG_ZOMBIE;
 652                         else if (p->exit_state == EXIT_DEAD)
 653                                 status = LTTNG_DEAD;
 654                         else if (lttng_task_is_running(p)) {
 655                                 /* Is this a forked child that has not run yet? */
 656                                 if (list_empty(&p->rt.run_list))
 657                                         status = LTTNG_WAIT_FORK;
 658                                 else
 659                                         /*
 660                                          * All tasks are considered as wait_cpu;
 661                                          * the viewer will sort out if the task
 662                                          * was really running at this time.
 663                                          */
 664                                         status = LTTNG_WAIT_CPU;
 665                         } else if (lttng_get_task_state(p) &
 666                                 (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)) {
 667                                 /* Task is waiting for something to complete */
 668                                 status = LTTNG_WAIT;
 669                         } else
 670                                 status = LTTNG_UNNAMED;
 671                         submode = LTTNG_NONE;
 672
 673                         /*
 674                          * Verification of t->mm is to filter out kernel
 675                          * threads; Viewer will further filter out if a
 676                          * user-space thread was in syscall mode or not.
 677                          */
 678                         if (p->mm)
 679                                 type = LTTNG_USER_THREAD;
 680                         else
 681                                 type = LTTNG_KERNEL_THREAD;
 682                         files = p->files;
 683
 684                         trace_lttng_statedump_process_state(session,
 685                                 p, type, mode, submode, status, files);
 686                         lttng_statedump_process_ns(session,
 687                                 p, type, mode, submode, status);
 688                         /*
 689                          * As an optimisation for the common case, do not
 690                          * repeat information for the same files_struct in
 691                          * two consecutive threads. This is the common case
 692                          * for threads sharing the same fd table. RCU guarantees
 693                          * that the same files_struct pointer is not re-used
 694                          * throughout processes/threads iteration.
 695                          */
 696                         if (files && files != prev_files) {
 697                                 lttng_enumerate_files(session, files, tmp);
 698                                 prev_files = files;
 699                         }
 700                         task_unlock(p);
 701                 } while_each_thread(g, p);
 702         }
 703         rcu_read_unlock();
 704
 705         free_page((unsigned long) tmp);
 706
 707         return 0;
 708 }
 709
 710 static
 711 void lttng_statedump_work_func(struct work_struct *work)
 712 {
 713         if (atomic_dec_and_test(&kernel_threads_to_run))
 714                 /* If we are the last thread, wake up do_lttng_statedump */
 715                 wake_up(&statedump_wq);
 716 }
 717
 718 static
 719 int do_lttng_statedump(struct lttng_session *session)
 720 {
 721         int cpu, ret;
 722
 723         trace_lttng_statedump_start(session);
 724         ret = lttng_enumerate_process_states(session);
 725         if (ret)
 726                 return ret;
 727         /*
 728          * FIXME
 729          * ret = lttng_enumerate_vm_maps(session);
 730          * if (ret)
 731          *      return ret;
 732          */
 733         ret = lttng_list_interrupts(session);
 734         if (ret)
 735                 return ret;
 736         ret = lttng_enumerate_network_ip_interface(session);
 737         if (ret)
 738                 return ret;
 739         ret = lttng_enumerate_block_devices(session);
 740         switch (ret) {
 741         case 0:
 742                 break;
 743         case -ENOSYS:
 744                 printk(KERN_WARNING "LTTng: block device enumeration is not supported by kernel\n");
 745                 break;
 746         default:
 747                 return ret;
 748         }
 749         ret = lttng_enumerate_cpu_topology(session);
 750         if (ret)
 751                 return ret;
 752
 753         /* TODO lttng_dump_idt_table(session); */
 754         /* TODO lttng_dump_softirq_vec(session); */
 755         /* TODO lttng_list_modules(session); */
 756         /* TODO lttng_dump_swap_files(session); */
 757
 758         /*
 759          * Fire off a work queue on each CPU. Their sole purpose in life
 760          * is to guarantee that each CPU has been in a state where is was in
 761          * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
 762          */
 763         get_online_cpus();
 764         atomic_set(&kernel_threads_to_run, num_online_cpus());
 765         for_each_online_cpu(cpu) {
 766                 INIT_DELAYED_WORK(&cpu_work[cpu], lttng_statedump_work_func);
 767                 schedule_delayed_work_on(cpu, &cpu_work[cpu], 0);
 768         }
 769         /* Wait for all threads to run */
 770         __wait_event(statedump_wq, (atomic_read(&kernel_threads_to_run) == 0));
 771         put_online_cpus();
 772         /* Our work is done */
 773         trace_lttng_statedump_end(session);
 774         return 0;
 775 }
 776
 777 /*
 778  * Called with session mutex held.
 779  */
 780 int lttng_statedump_start(struct lttng_session *session)
 781 {
 782         return do_lttng_statedump(session);
 783 }
 784 EXPORT_SYMBOL_GPL(lttng_statedump_start);
 785
 786 static
 787 int __init lttng_statedump_init(void)
 788 {
 789         /*
 790          * Allow module to load even if the fixup cannot be done. This
 791          * will allow seemless transition when the underlying issue fix
 792          * is merged into the Linux kernel, and when tracepoint.c
 793          * "tracepoint_module_notify" is turned into a static function.
 794          */
 795         (void) wrapper_lttng_fixup_sig(THIS_MODULE);
 796         return 0;
 797 }
 798
 799 module_init(lttng_statedump_init);
 800
 801 static
 802 void __exit lttng_statedump_exit(void)
 803 {
 804 }
 805
 806 module_exit(lttng_statedump_exit);
 807
 808 MODULE_LICENSE("GPL and additional rights");
 809 MODULE_AUTHOR("Jean-Hugues Deschenes");
 810 MODULE_DESCRIPTION("LTTng statedump provider");
 811 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION) "."
 812         __stringify(LTTNG_MODULES_MINOR_VERSION) "."
 813         __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION)
 814         LTTNG_MODULES_EXTRAVERSION);