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