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