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/netdevice.h>
  27 #include <linux/inetdevice.h>
  28 #include <linux/mm.h>
  29 #include <linux/swap.h>
  30 #include <linux/wait.h>
  31 #include <linux/mutex.h>
  32 #include <linux/device.h>
  33
  34 #include <linux/blkdev.h>
  35
  36 #include <lttng/events.h>
  37 #include <lttng/tracer.h>
  38 #include <lttng/events-internal.h>
  39 #include <wrapper/cpu.h>
  40 #include <wrapper/irqdesc.h>
  41 #include <wrapper/fdtable.h>
  42 #include <wrapper/namespace.h>
  43 #include <wrapper/irq.h>
  44 #include <wrapper/tracepoint.h>
  45 #include <wrapper/genhd.h>
  46 #include <wrapper/file.h>
  47 #include <wrapper/fdtable.h>
  48 #include <wrapper/sched.h>
  49
  50 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
  51 #include <linux/irq.h>
  52 #endif
  53
  54 /* Define the tracepoints, but do not build the probes */
  55 #define CREATE_TRACE_POINTS
  56 #define TRACE_INCLUDE_PATH instrumentation/events
  57 #define TRACE_INCLUDE_FILE lttng-statedump
  58 #define LTTNG_INSTRUMENTATION
  59 #include <instrumentation/events/lttng-statedump.h>
  60
  61 LTTNG_DEFINE_TRACE(lttng_statedump_block_device,
  62         TP_PROTO(struct lttng_kernel_session *session,
  63                 dev_t dev, const char *diskname),
  64         TP_ARGS(session, dev, diskname));
  65
  66 LTTNG_DEFINE_TRACE(lttng_statedump_end,
  67         TP_PROTO(struct lttng_kernel_session *session),
  68         TP_ARGS(session));
  69
  70 LTTNG_DEFINE_TRACE(lttng_statedump_interrupt,
  71         TP_PROTO(struct lttng_kernel_session *session,
  72                 unsigned int irq, const char *chip_name,
  73                 struct irqaction *action),
  74         TP_ARGS(session, irq, chip_name, action));
  75
  76 LTTNG_DEFINE_TRACE(lttng_statedump_file_descriptor,
  77         TP_PROTO(struct lttng_kernel_session *session,
  78                 struct files_struct *files,
  79                 int fd, const char *filename,
  80                 unsigned int flags, fmode_t fmode),
  81         TP_ARGS(session, files, fd, filename, flags, fmode));
  82
  83 LTTNG_DEFINE_TRACE(lttng_statedump_start,
  84         TP_PROTO(struct lttng_kernel_session *session),
  85         TP_ARGS(session));
  86
  87 LTTNG_DEFINE_TRACE(lttng_statedump_process_state,
  88         TP_PROTO(struct lttng_kernel_session *session,
  89                 struct task_struct *p,
  90                 int type, int mode, int submode, int status,
  91                 struct files_struct *files),
  92         TP_ARGS(session, p, type, mode, submode, status, files));
  93
  94 LTTNG_DEFINE_TRACE(lttng_statedump_process_pid_ns,
  95         TP_PROTO(struct lttng_kernel_session *session,
  96                 struct task_struct *p,
  97                 struct pid_namespace *pid_ns),
  98         TP_ARGS(session, p, pid_ns));
  99
 100 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(4,6,0))
 101 LTTNG_DEFINE_TRACE(lttng_statedump_process_cgroup_ns,
 102         TP_PROTO(struct lttng_kernel_session *session,
 103                 struct task_struct *p,
 104                 struct cgroup_namespace *cgroup_ns),
 105         TP_ARGS(session, p, cgroup_ns));
 106 #endif
 107
 108 LTTNG_DEFINE_TRACE(lttng_statedump_process_ipc_ns,
 109         TP_PROTO(struct lttng_kernel_session *session,
 110                 struct task_struct *p,
 111                 struct ipc_namespace *ipc_ns),
 112         TP_ARGS(session, p, ipc_ns));
 113
 114 #ifndef LTTNG_MNT_NS_MISSING_HEADER
 115 LTTNG_DEFINE_TRACE(lttng_statedump_process_mnt_ns,
 116         TP_PROTO(struct lttng_kernel_session *session,
 117                 struct task_struct *p,
 118                 struct mnt_namespace *mnt_ns),
 119         TP_ARGS(session, p, mnt_ns));
 120 #endif
 121
 122 LTTNG_DEFINE_TRACE(lttng_statedump_process_net_ns,
 123         TP_PROTO(struct lttng_kernel_session *session,
 124                 struct task_struct *p,
 125                 struct net *net_ns),
 126         TP_ARGS(session, p, net_ns));
 127
 128 LTTNG_DEFINE_TRACE(lttng_statedump_process_user_ns,
 129         TP_PROTO(struct lttng_kernel_session *session,
 130                 struct task_struct *p,
 131                 struct user_namespace *user_ns),
 132         TP_ARGS(session, p, user_ns));
 133
 134 LTTNG_DEFINE_TRACE(lttng_statedump_process_uts_ns,
 135         TP_PROTO(struct lttng_kernel_session *session,
 136                 struct task_struct *p,
 137                 struct uts_namespace *uts_ns),
 138         TP_ARGS(session, p, uts_ns));
 139
 140 LTTNG_DEFINE_TRACE(lttng_statedump_process_time_ns,
 141         TP_PROTO(struct lttng_kernel_session *session,
 142                 struct task_struct *p,
 143                 struct time_namespace *time_ns),
 144         TP_ARGS(session, p, time_ns));
 145
 146 LTTNG_DEFINE_TRACE(lttng_statedump_network_interface,
 147         TP_PROTO(struct lttng_kernel_session *session,
 148                 struct net_device *dev, struct in_ifaddr *ifa),
 149         TP_ARGS(session, dev, ifa));
 150
 151 #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
 152 LTTNG_DEFINE_TRACE(lttng_statedump_cpu_topology,
 153         TP_PROTO(struct lttng_kernel_session *session, struct cpuinfo_x86 *c),
 154         TP_ARGS(session, c));
 155 #endif
 156
 157 struct lttng_fd_ctx {
 158         char *page;
 159         struct lttng_kernel_session *session;
 160         struct files_struct *files;
 161 };
 162
 163 /*
 164  * Protected by the trace lock.
 165  */
 166 static struct delayed_work cpu_work[NR_CPUS];
 167 static DECLARE_WAIT_QUEUE_HEAD(statedump_wq);
 168 static atomic_t kernel_threads_to_run;
 169
 170 enum lttng_thread_type {
 171         LTTNG_USER_THREAD = 0,
 172         LTTNG_KERNEL_THREAD = 1,
 173 };
 174
 175 enum lttng_execution_mode {
 176         LTTNG_USER_MODE = 0,
 177         LTTNG_SYSCALL = 1,
 178         LTTNG_TRAP = 2,
 179         LTTNG_IRQ = 3,
 180         LTTNG_SOFTIRQ = 4,
 181         LTTNG_MODE_UNKNOWN = 5,
 182 };
 183
 184 enum lttng_execution_submode {
 185         LTTNG_NONE = 0,
 186         LTTNG_UNKNOWN = 1,
 187 };
 188
 189 enum lttng_process_status {
 190         LTTNG_UNNAMED = 0,
 191         LTTNG_WAIT_FORK = 1,
 192         LTTNG_WAIT_CPU = 2,
 193         LTTNG_EXIT = 3,
 194         LTTNG_ZOMBIE = 4,
 195         LTTNG_WAIT = 5,
 196         LTTNG_RUN = 6,
 197         LTTNG_DEAD = 7,
 198 };
 199
 200
 201 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(6,0,0) || \
 202     LTTNG_RHEL_KERNEL_RANGE(5,14,0,163,0,0, 5,15,0,0,0,0))
 203
 204 #define LTTNG_PART_STRUCT_TYPE struct block_device
 205
 206 static
 207 int lttng_get_part_name(struct gendisk *disk, struct block_device *part, char *name_buf)
 208 {
 209         int ret;
 210
 211         ret = snprintf(name_buf, BDEVNAME_SIZE, "%pg", part);
 212         if (ret < 0 || ret >= BDEVNAME_SIZE)
 213                 return -ENOSYS;
 214
 215         return 0;
 216 }
 217
 218 static
 219 dev_t lttng_get_part_devt(struct block_device *part)
 220 {
 221         return part->bd_dev;
 222 }
 223
 224 #elif (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,11,0))
 225
 226 #define LTTNG_PART_STRUCT_TYPE struct block_device
 227
 228 static
 229 int lttng_get_part_name(struct gendisk *disk, struct block_device *part, char *name_buf)
 230 {
 231         const char *p;
 232
 233         p = bdevname(part, name_buf);
 234         if (!p)
 235                 return -ENOSYS;
 236
 237         return 0;
 238 }
 239
 240 static
 241 dev_t lttng_get_part_devt(struct block_device *part)
 242 {
 243         return part->bd_dev;
 244 }
 245
 246 #else
 247
 248 #define LTTNG_PART_STRUCT_TYPE struct hd_struct
 249
 250 static
 251 int lttng_get_part_name(struct gendisk *disk, struct hd_struct *part, char *name_buf)
 252 {
 253         const char *p;
 254         struct block_device bdev;
 255
 256         /*
 257          * Create a partial 'struct blockdevice' to use
 258          * 'bdevname()' which is a simple wrapper over
 259          * 'disk_name()' but has the honor to be EXPORT_SYMBOL.
 260          */
 261         bdev.bd_disk = disk;
 262         bdev.bd_part = part;
 263
 264         p = bdevname(&bdev, name_buf);
 265         if (!p)
 266                 return -ENOSYS;
 267
 268         return 0;
 269 }
 270
 271 static
 272 dev_t lttng_get_part_devt(struct hd_struct *part)
 273 {
 274         return part_devt(part);
 275 }
 276 #endif
 277
 278 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,12,0))
 279 static
 280 int lttng_statedump_each_block_device(struct lttng_kernel_session *session, struct gendisk *disk)
 281 {
 282         struct block_device *part;
 283         unsigned long idx;
 284         int ret = 0;
 285
 286         /* Include partition 0 */
 287         idx = 0;
 288
 289         rcu_read_lock();
 290         xa_for_each(&disk->part_tbl, idx, part) {
 291                 char name_buf[BDEVNAME_SIZE];
 292
 293                 /* Exclude non-partitions bdev and empty partitions. */
 294                 if (bdev_is_partition(part) && !bdev_nr_sectors(part))
 295                         continue;
 296
 297                 if (lttng_get_part_name(disk, part, name_buf) == -ENOSYS) {
 298                         ret = -ENOSYS;
 299                         goto end;
 300                 }
 301                 trace_lttng_statedump_block_device(session, lttng_get_part_devt(part),
 302                                 name_buf);
 303         }
 304 end:
 305         rcu_read_unlock();
 306         return ret;
 307 }
 308 #else
 309 static
 310 int lttng_statedump_each_block_device(struct lttng_kernel_session *session, struct gendisk *disk)
 311 {
 312         struct disk_part_iter piter;
 313         LTTNG_PART_STRUCT_TYPE *part;
 314
 315         disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
 316
 317         while ((part = disk_part_iter_next(&piter))) {
 318                 char name_buf[BDEVNAME_SIZE];
 319
 320                 if (lttng_get_part_name(disk, part, name_buf) == -ENOSYS) {
 321                         disk_part_iter_exit(&piter);
 322                         return -ENOSYS;
 323                 }
 324                 trace_lttng_statedump_block_device(session, lttng_get_part_devt(part),
 325                                 name_buf);
 326         }
 327         disk_part_iter_exit(&piter);
 328
 329         return 0;
 330 }
 331 #endif
 332
 333 static
 334 int lttng_enumerate_block_devices(struct lttng_kernel_session *session)
 335 {
 336         struct class *ptr_block_class;
 337         struct device_type *ptr_disk_type;
 338         struct class_dev_iter iter;
 339         struct device *dev;
 340         int ret = 0;
 341
 342         ptr_block_class = wrapper_get_block_class();
 343         if (!ptr_block_class) {
 344                 ret = -ENOSYS;
 345                 goto end;
 346         }
 347         ptr_disk_type = wrapper_get_disk_type();
 348         if (!ptr_disk_type) {
 349                 ret = -ENOSYS;
 350                 goto end;
 351         }
 352         class_dev_iter_init(&iter, ptr_block_class, NULL, ptr_disk_type);
 353         while ((dev = class_dev_iter_next(&iter))) {
 354                 struct gendisk *disk = dev_to_disk(dev);
 355
 356                 /*
 357                  * Don't show empty devices or things that have been
 358                  * suppressed
 359                  */
 360                 if (get_capacity(disk) == 0 ||
 361                     (disk->flags & LTTNG_GENHD_FL_HIDDEN))
 362                         continue;
 363
 364                 ret = lttng_statedump_each_block_device(session, disk);
 365         }
 366         class_dev_iter_exit(&iter);
 367 end:
 368         return ret;
 369 }
 370
 371 #ifdef CONFIG_INET
 372
 373 static
 374 void lttng_enumerate_device(struct lttng_kernel_session *session,
 375                 struct net_device *dev)
 376 {
 377         struct in_device *in_dev;
 378         struct in_ifaddr *ifa;
 379
 380         if (dev->flags & IFF_UP) {
 381                 in_dev = in_dev_get(dev);
 382                 if (in_dev) {
 383                         for (ifa = in_dev->ifa_list; ifa != NULL;
 384                              ifa = ifa->ifa_next) {
 385                                 trace_lttng_statedump_network_interface(
 386                                         session, dev, ifa);
 387                         }
 388                         in_dev_put(in_dev);
 389                 }
 390         } else {
 391                 trace_lttng_statedump_network_interface(
 392                         session, dev, NULL);
 393         }
 394 }
 395
 396 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(6,9,0))
 397 static
 398 int lttng_enumerate_network_ip_interface(struct lttng_kernel_session *session)
 399 {
 400         struct net_device *dev;
 401
 402         rtnl_lock();
 403         for_each_netdev(&init_net, dev)
 404                 lttng_enumerate_device(session, dev);
 405         rtnl_unlock();
 406
 407         return 0;
 408 }
 409 #else
 410 static
 411 int lttng_enumerate_network_ip_interface(struct lttng_kernel_session *session)
 412 {
 413         struct net_device *dev;
 414
 415         read_lock(&dev_base_lock);
 416         for_each_netdev(&init_net, dev)
 417                 lttng_enumerate_device(session, dev);
 418         read_unlock(&dev_base_lock);
 419
 420         return 0;
 421 }
 422 #endif /* (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(6,9,0)) */
 423 #else /* CONFIG_INET */
 424 static inline
 425 int lttng_enumerate_network_ip_interface(struct lttng_kernel_session *session)
 426 {
 427         return 0;
 428 }
 429 #endif /* CONFIG_INET */
 430
 431 static
 432 int lttng_dump_one_fd(const void *p, struct file *file, unsigned int fd)
 433 {
 434         const struct lttng_fd_ctx *ctx = p;
 435         const char *s = d_path(&file->f_path, ctx->page, PAGE_SIZE);
 436         unsigned int flags = file->f_flags;
 437         struct fdtable *fdt;
 438
 439         /*
 440          * We don't expose kernel internal flags, only userspace-visible
 441          * flags.
 442          */
 443         flags &= ~FMODE_NONOTIFY;
 444         fdt = files_fdtable(ctx->files);
 445         /*
 446          * We need to check here again whether fd is within the fdt
 447          * max_fds range, because we might be seeing a different
 448          * files_fdtable() than iterate_fd(), assuming only RCU is
 449          * protecting the read. In reality, iterate_fd() holds
 450          * file_lock, which should ensure the fdt does not change while
 451          * the lock is taken, but we are not aware whether this is
 452          * guaranteed or not, so play safe.
 453          */
 454         if (fd < fdt->max_fds && lttng_close_on_exec(fd, fdt))
 455                 flags |= O_CLOEXEC;
 456         if (IS_ERR(s)) {
 457                 struct dentry *dentry = file->f_path.dentry;
 458
 459                 /* Make sure we give at least some info */
 460                 spin_lock(&dentry->d_lock);
 461                 trace_lttng_statedump_file_descriptor(ctx->session,
 462                         ctx->files, fd, dentry->d_name.name, flags,
 463                         file->f_mode);
 464                 spin_unlock(&dentry->d_lock);
 465                 goto end;
 466         }
 467         trace_lttng_statedump_file_descriptor(ctx->session,
 468                 ctx->files, fd, s, flags, file->f_mode);
 469 end:
 470         return 0;
 471 }
 472
 473 /* Called with task lock held. */
 474 static
 475 void lttng_enumerate_files(struct lttng_kernel_session *session,
 476                 struct files_struct *files,
 477                 char *tmp)
 478 {
 479         struct lttng_fd_ctx ctx = { .page = tmp, .session = session, .files = files, };
 480
 481         lttng_iterate_fd(files, 0, lttng_dump_one_fd, &ctx);
 482 }
 483
 484 #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
 485 static
 486 int lttng_enumerate_cpu_topology(struct lttng_kernel_session *session)
 487 {
 488         int cpu;
 489         const cpumask_t *cpumask = cpu_possible_mask;
 490
 491         for (cpu = cpumask_first(cpumask); cpu < nr_cpu_ids;
 492                         cpu = cpumask_next(cpu, cpumask)) {
 493                 trace_lttng_statedump_cpu_topology(session, &cpu_data(cpu));
 494         }
 495
 496         return 0;
 497 }
 498 #else
 499 static
 500 int lttng_enumerate_cpu_topology(struct lttng_kernel_session *session)
 501 {
 502         return 0;
 503 }
 504 #endif
 505
 506 #if 0
 507 /*
 508  * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
 509  * (scheduling in atomic). Normally, the tasklist lock protects this kind of
 510  * iteration, but it is not exported to modules.
 511  */
 512 static
 513 void lttng_enumerate_task_vm_maps(struct lttng_kernel_session *session,
 514                 struct task_struct *p)
 515 {
 516         struct mm_struct *mm;
 517         struct vm_area_struct *map;
 518         unsigned long ino;
 519
 520         /* get_task_mm does a task_lock... */
 521         mm = get_task_mm(p);
 522         if (!mm)
 523                 return;
 524
 525         map = mm->mmap;
 526         if (map) {
 527                 down_read(&mm->mmap_sem);
 528                 while (map) {
 529                         if (map->vm_file)
 530                                 ino = map->vm_file->lttng_f_dentry->d_inode->i_ino;
 531                         else
 532                                 ino = 0;
 533                         trace_lttng_statedump_vm_map(session, p, map, ino);
 534                         map = map->vm_next;
 535                 }
 536                 up_read(&mm->mmap_sem);
 537         }
 538         mmput(mm);
 539 }
 540
 541 static
 542 int lttng_enumerate_vm_maps(struct lttng_kernel_session *session)
 543 {
 544         struct task_struct *p;
 545
 546         rcu_read_lock();
 547         for_each_process(p)
 548                 lttng_enumerate_task_vm_maps(session, p);
 549         rcu_read_unlock();
 550         return 0;
 551 }
 552 #endif
 553
 554 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
 555
 556 static
 557 int lttng_list_interrupts(struct lttng_kernel_session *session)
 558 {
 559         unsigned int irq;
 560         unsigned long flags = 0;
 561         struct irq_desc *desc;
 562
 563 #define irq_to_desc     wrapper_irq_to_desc
 564         /* needs irq_desc */
 565         for_each_irq_desc(irq, desc) {
 566                 struct irqaction *action;
 567                 const char *irq_chip_name =
 568                         irq_desc_get_chip(desc)->name ? : "unnamed_irq_chip";
 569
 570                 local_irq_save(flags);
 571                 raw_spin_lock(&desc->lock);
 572                 for (action = desc->action; action; action = action->next) {
 573                         trace_lttng_statedump_interrupt(session,
 574                                 irq, irq_chip_name, action);
 575                 }
 576                 raw_spin_unlock(&desc->lock);
 577                 local_irq_restore(flags);
 578         }
 579         return 0;
 580 #undef irq_to_desc
 581 }
 582 #else
 583 static inline
 584 int lttng_list_interrupts(struct lttng_kernel_session *session)
 585 {
 586         return 0;
 587 }
 588 #endif
 589
 590 /*
 591  * Statedump the task's namespaces using the proc filesystem inode number as
 592  * the unique identifier. The user and pid ns are nested and will be dumped
 593  * recursively.
 594  *
 595  * Called with task lock held.
 596  */
 597 static
 598 void lttng_statedump_process_ns(struct lttng_kernel_session *session,
 599                 struct task_struct *p,
 600                 enum lttng_thread_type type,
 601                 enum lttng_execution_mode mode,
 602                 enum lttng_execution_submode submode,
 603                 enum lttng_process_status status)
 604 {
 605         struct nsproxy *proxy;
 606         struct pid_namespace *pid_ns;
 607         struct user_namespace *user_ns;
 608
 609         /*
 610          * The pid and user namespaces are special, they are nested and
 611          * accessed with specific functions instead of the nsproxy struct
 612          * like the other namespaces.
 613          */
 614         pid_ns = task_active_pid_ns(p);
 615         do {
 616                 trace_lttng_statedump_process_pid_ns(session, p, pid_ns);
 617                 pid_ns = pid_ns ? pid_ns->parent : NULL;
 618         } while (pid_ns);
 619
 620
 621         user_ns = task_cred_xxx(p, user_ns);
 622         do {
 623                 trace_lttng_statedump_process_user_ns(session, p, user_ns);
 624                 /*
 625                  * trace_lttng_statedump_process_user_ns() internally
 626                  * checks whether user_ns is NULL. While this does not
 627                  * appear to be a possible return value for
 628                  * task_cred_xxx(), err on the safe side and check
 629                  * for NULL here as well to be consistent with the
 630                  * paranoid behavior of
 631                  * trace_lttng_statedump_process_user_ns().
 632                  */
 633                 user_ns = user_ns ? user_ns->lttng_user_ns_parent : NULL;
 634         } while (user_ns);
 635
 636         /*
 637          * Back and forth on locking strategy within Linux upstream for nsproxy.
 638          * See Linux upstream commit 728dba3a39c66b3d8ac889ddbe38b5b1c264aec3
 639          * "namespaces: Use task_lock and not rcu to protect nsproxy"
 640          * for details.
 641          */
 642 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(3,17,0) || \
 643                 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
 644                 LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \
 645                 LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
 646         proxy = p->nsproxy;
 647 #else
 648         rcu_read_lock();
 649         proxy = task_nsproxy(p);
 650 #endif
 651         if (proxy) {
 652 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(4,6,0))
 653                 trace_lttng_statedump_process_cgroup_ns(session, p, proxy->cgroup_ns);
 654 #endif
 655                 trace_lttng_statedump_process_ipc_ns(session, p, proxy->ipc_ns);
 656 #ifndef LTTNG_MNT_NS_MISSING_HEADER
 657                 trace_lttng_statedump_process_mnt_ns(session, p, proxy->mnt_ns);
 658 #endif
 659                 trace_lttng_statedump_process_net_ns(session, p, proxy->net_ns);
 660                 trace_lttng_statedump_process_uts_ns(session, p, proxy->uts_ns);
 661 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,6,0) || \
 662         LTTNG_RHEL_KERNEL_RANGE(4,18,0,305,0,0, 4,19,0,0,0,0))
 663                 trace_lttng_statedump_process_time_ns(session, p, proxy->time_ns);
 664 #endif
 665         }
 666 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(3,17,0) || \
 667                 LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
 668                 LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \
 669                 LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
 670         /* (nothing) */
 671 #else
 672         rcu_read_unlock();
 673 #endif
 674 }
 675
 676 static
 677 int lttng_enumerate_process_states(struct lttng_kernel_session *session)
 678 {
 679         struct task_struct *g, *p;
 680         char *tmp;
 681
 682         tmp = (char *) __get_free_page(GFP_KERNEL);
 683         if (!tmp)
 684                 return -ENOMEM;
 685
 686         rcu_read_lock();
 687         for_each_process(g) {
 688                 struct files_struct *prev_files = NULL;
 689
 690                 p = g;
 691                 do {
 692                         enum lttng_execution_mode mode =
 693                                 LTTNG_MODE_UNKNOWN;
 694                         enum lttng_execution_submode submode =
 695                                 LTTNG_UNKNOWN;
 696                         enum lttng_process_status status;
 697                         enum lttng_thread_type type;
 698                         struct files_struct *files;
 699
 700                         task_lock(p);
 701                         if (p->exit_state == EXIT_ZOMBIE)
 702                                 status = LTTNG_ZOMBIE;
 703                         else if (p->exit_state == EXIT_DEAD)
 704                                 status = LTTNG_DEAD;
 705                         else if (lttng_task_is_running(p)) {
 706                                 /* Is this a forked child that has not run yet? */
 707                                 if (list_empty(&p->rt.run_list))
 708                                         status = LTTNG_WAIT_FORK;
 709                                 else
 710                                         /*
 711                                          * All tasks are considered as wait_cpu;
 712                                          * the viewer will sort out if the task
 713                                          * was really running at this time.
 714                                          */
 715                                         status = LTTNG_WAIT_CPU;
 716                         } else if (lttng_get_task_state(p) &
 717                                 (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)) {
 718                                 /* Task is waiting for something to complete */
 719                                 status = LTTNG_WAIT;
 720                         } else
 721                                 status = LTTNG_UNNAMED;
 722                         submode = LTTNG_NONE;
 723
 724                         /*
 725                          * Verification of t->mm is to filter out kernel
 726                          * threads; Viewer will further filter out if a
 727                          * user-space thread was in syscall mode or not.
 728                          */
 729                         if (p->mm)
 730                                 type = LTTNG_USER_THREAD;
 731                         else
 732                                 type = LTTNG_KERNEL_THREAD;
 733                         files = p->files;
 734
 735                         trace_lttng_statedump_process_state(session,
 736                                 p, type, mode, submode, status, files);
 737                         lttng_statedump_process_ns(session,
 738                                 p, type, mode, submode, status);
 739                         /*
 740                          * As an optimisation for the common case, do not
 741                          * repeat information for the same files_struct in
 742                          * two consecutive threads. This is the common case
 743                          * for threads sharing the same fd table. RCU guarantees
 744                          * that the same files_struct pointer is not re-used
 745                          * throughout processes/threads iteration.
 746                          */
 747                         if (files && files != prev_files) {
 748                                 lttng_enumerate_files(session, files, tmp);
 749                                 prev_files = files;
 750                         }
 751                         task_unlock(p);
 752                 } while_each_thread(g, p);
 753         }
 754         rcu_read_unlock();
 755
 756         free_page((unsigned long) tmp);
 757
 758         return 0;
 759 }
 760
 761 static
 762 void lttng_statedump_work_func(struct work_struct *work)
 763 {
 764         if (atomic_dec_and_test(&kernel_threads_to_run))
 765                 /* If we are the last thread, wake up do_lttng_statedump */
 766                 wake_up(&statedump_wq);
 767 }
 768
 769 static
 770 int do_lttng_statedump(struct lttng_kernel_session *session)
 771 {
 772         int cpu, ret;
 773
 774         trace_lttng_statedump_start(session);
 775         ret = lttng_enumerate_process_states(session);
 776         if (ret)
 777                 return ret;
 778         /*
 779          * FIXME
 780          * ret = lttng_enumerate_vm_maps(session);
 781          * if (ret)
 782          *      return ret;
 783          */
 784         ret = lttng_list_interrupts(session);
 785         if (ret)
 786                 return ret;
 787         ret = lttng_enumerate_network_ip_interface(session);
 788         if (ret)
 789                 return ret;
 790         ret = lttng_enumerate_block_devices(session);
 791         switch (ret) {
 792         case 0:
 793                 break;
 794         case -ENOSYS:
 795                 printk(KERN_WARNING "LTTng: block device enumeration is not supported by kernel\n");
 796                 break;
 797         default:
 798                 return ret;
 799         }
 800         ret = lttng_enumerate_cpu_topology(session);
 801         if (ret)
 802                 return ret;
 803
 804         /* TODO lttng_dump_idt_table(session); */
 805         /* TODO lttng_dump_softirq_vec(session); */
 806         /* TODO lttng_list_modules(session); */
 807         /* TODO lttng_dump_swap_files(session); */
 808
 809         /*
 810          * Fire off a work queue on each CPU. Their sole purpose in life
 811          * is to guarantee that each CPU has been in a state where is was in
 812          * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
 813          */
 814         lttng_cpus_read_lock();
 815         atomic_set(&kernel_threads_to_run, num_online_cpus());
 816         for_each_online_cpu(cpu) {
 817                 INIT_DELAYED_WORK(&cpu_work[cpu], lttng_statedump_work_func);
 818                 schedule_delayed_work_on(cpu, &cpu_work[cpu], 0);
 819         }
 820         /* Wait for all threads to run */
 821         __wait_event(statedump_wq, (atomic_read(&kernel_threads_to_run) == 0));
 822         lttng_cpus_read_unlock();
 823         /* Our work is done */
 824         trace_lttng_statedump_end(session);
 825         return 0;
 826 }
 827
 828 /*
 829  * Called with session mutex held.
 830  */
 831 int lttng_statedump_start(struct lttng_kernel_session *session)
 832 {
 833         return do_lttng_statedump(session);
 834 }
 835 EXPORT_SYMBOL_GPL(lttng_statedump_start);
 836
 837 static
 838 int __init lttng_statedump_init(void)
 839 {
 840         /*
 841          * Allow module to load even if the fixup cannot be done. This
 842          * will allow seemless transition when the underlying issue fix
 843          * is merged into the Linux kernel, and when tracepoint.c
 844          * "tracepoint_module_notify" is turned into a static function.
 845          */
 846         (void) wrapper_lttng_fixup_sig(THIS_MODULE);
 847         return 0;
 848 }
 849
 850 module_init(lttng_statedump_init);
 851
 852 static
 853 void __exit lttng_statedump_exit(void)
 854 {
 855 }
 856
 857 module_exit(lttng_statedump_exit);
 858
 859 MODULE_LICENSE("GPL and additional rights");
 860 MODULE_AUTHOR("Jean-Hugues Deschenes");
 861 MODULE_DESCRIPTION("LTTng statedump provider");
 862 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION) "."
 863         __stringify(LTTNG_MODULES_MINOR_VERSION) "."
 864         __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION)
 865         LTTNG_MODULES_EXTRAVERSION);