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