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