lttng-statedump-impl.c

   1 /*
   2  * lttng-statedump.c
   3  *
   4  * Linux Trace Toolkit Next Generation Kernel State Dump
   5  *
   6  * Copyright 2005 Jean-Hugues Deschenes <jean-hugues.deschenes@polymtl.ca>
   7  * Copyright 2006-2012 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   8  *
   9  * This library is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public
  11  * License as published by the Free Software Foundation; only
  12  * version 2.1 of the License.
  13  *
  14  * This library is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with this library; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  22  *
  23  * Changes:
  24  *      Eric Clement:                   Add listing of network IP interface
  25  *      2006, 2007 Mathieu Desnoyers    Fix kernel threads
  26  *                                      Various updates
  27  */
  28
  29 #include <linux/init.h>
  30 #include <linux/module.h>
  31 #include <linux/netlink.h>
  32 #include <linux/inet.h>
  33 #include <linux/ip.h>
  34 #include <linux/kthread.h>
  35 #include <linux/proc_fs.h>
  36 #include <linux/file.h>
  37 #include <linux/interrupt.h>
  38 #include <linux/irqnr.h>
  39 #include <linux/cpu.h>
  40 #include <linux/netdevice.h>
  41 #include <linux/inetdevice.h>
  42 #include <linux/sched.h>
  43 #include <linux/mm.h>
  44 #include <linux/fdtable.h>
  45 #include <linux/swap.h>
  46 #include <linux/wait.h>
  47 #include <linux/mutex.h>
  48 #include <linux/device.h>
  49
  50 #include "lttng-events.h"
  51 #include "lttng-tracer.h"
  52 #include "wrapper/irqdesc.h"
  53 #include "wrapper/spinlock.h"
  54 #include "wrapper/fdtable.h"
  55 #include "wrapper/nsproxy.h"
  56 #include "wrapper/irq.h"
  57 #include "wrapper/tracepoint.h"
  58 #include "wrapper/genhd.h"
  59
  60 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
  61 #include <linux/irq.h>
  62 #endif
  63
  64 /* Define the tracepoints, but do not build the probes */
  65 #define CREATE_TRACE_POINTS
  66 #define TRACE_INCLUDE_PATH ../instrumentation/events/lttng-module
  67 #define TRACE_INCLUDE_FILE lttng-statedump
  68 #include "instrumentation/events/lttng-module/lttng-statedump.h"
  69
  70 DEFINE_TRACE(lttng_statedump_block_device);
  71 DEFINE_TRACE(lttng_statedump_end);
  72 DEFINE_TRACE(lttng_statedump_interrupt);
  73 DEFINE_TRACE(lttng_statedump_file_descriptor);
  74 DEFINE_TRACE(lttng_statedump_start);
  75 DEFINE_TRACE(lttng_statedump_process_state);
  76 DEFINE_TRACE(lttng_statedump_network_interface);
  77
  78 struct lttng_fd_ctx {
  79         char *page;
  80         struct lttng_session *session;
  81         struct task_struct *p;
  82         struct fdtable *fdt;
  83 };
  84
  85 /*
  86  * Protected by the trace lock.
  87  */
  88 static struct delayed_work cpu_work[NR_CPUS];
  89 static DECLARE_WAIT_QUEUE_HEAD(statedump_wq);
  90 static atomic_t kernel_threads_to_run;
  91
  92 enum lttng_thread_type {
  93         LTTNG_USER_THREAD = 0,
  94         LTTNG_KERNEL_THREAD = 1,
  95 };
  96
  97 enum lttng_execution_mode {
  98         LTTNG_USER_MODE = 0,
  99         LTTNG_SYSCALL = 1,
 100         LTTNG_TRAP = 2,
 101         LTTNG_IRQ = 3,
 102         LTTNG_SOFTIRQ = 4,
 103         LTTNG_MODE_UNKNOWN = 5,
 104 };
 105
 106 enum lttng_execution_submode {
 107         LTTNG_NONE = 0,
 108         LTTNG_UNKNOWN = 1,
 109 };
 110
 111 enum lttng_process_status {
 112         LTTNG_UNNAMED = 0,
 113         LTTNG_WAIT_FORK = 1,
 114         LTTNG_WAIT_CPU = 2,
 115         LTTNG_EXIT = 3,
 116         LTTNG_ZOMBIE = 4,
 117         LTTNG_WAIT = 5,
 118         LTTNG_RUN = 6,
 119         LTTNG_DEAD = 7,
 120 };
 121
 122 static
 123 int lttng_enumerate_block_devices(struct lttng_session *session)
 124 {
 125         struct class *ptr_block_class;
 126         struct device_type *ptr_disk_type;
 127         struct class_dev_iter iter;
 128         struct device *dev;
 129
 130         ptr_block_class = wrapper_get_block_class();
 131         if (!ptr_block_class)
 132                 return -ENOSYS;
 133         ptr_disk_type = wrapper_get_disk_type();
 134         if (!ptr_disk_type) {
 135                 return -ENOSYS;
 136         }
 137         class_dev_iter_init(&iter, ptr_block_class, NULL, ptr_disk_type);
 138         while ((dev = class_dev_iter_next(&iter))) {
 139                 struct disk_part_iter piter;
 140                 struct gendisk *disk = dev_to_disk(dev);
 141                 struct hd_struct *part;
 142
 143                 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
 144                 while ((part = disk_part_iter_next(&piter))) {
 145                         char name_buf[BDEVNAME_SIZE];
 146                         char *p;
 147
 148                         p = wrapper_disk_name(disk, part->partno, name_buf);
 149                         if (!p) {
 150                                 disk_part_iter_exit(&piter);
 151                                 class_dev_iter_exit(&iter);
 152                                 return -ENOSYS;
 153                         }
 154                         trace_lttng_statedump_block_device(session,
 155                                         part_devt(part), name_buf);
 156                 }
 157                 disk_part_iter_exit(&piter);
 158         }
 159         class_dev_iter_exit(&iter);
 160         return 0;
 161 }
 162
 163 #ifdef CONFIG_INET
 164
 165 static
 166 void lttng_enumerate_device(struct lttng_session *session,
 167                 struct net_device *dev)
 168 {
 169         struct in_device *in_dev;
 170         struct in_ifaddr *ifa;
 171
 172         if (dev->flags & IFF_UP) {
 173                 in_dev = in_dev_get(dev);
 174                 if (in_dev) {
 175                         for (ifa = in_dev->ifa_list; ifa != NULL;
 176                              ifa = ifa->ifa_next) {
 177                                 trace_lttng_statedump_network_interface(
 178                                         session, dev, ifa);
 179                         }
 180                         in_dev_put(in_dev);
 181                 }
 182         } else {
 183                 trace_lttng_statedump_network_interface(
 184                         session, dev, NULL);
 185         }
 186 }
 187
 188 static
 189 int lttng_enumerate_network_ip_interface(struct lttng_session *session)
 190 {
 191         struct net_device *dev;
 192
 193         read_lock(&dev_base_lock);
 194         for_each_netdev(&init_net, dev)
 195                 lttng_enumerate_device(session, dev);
 196         read_unlock(&dev_base_lock);
 197
 198         return 0;
 199 }
 200 #else /* CONFIG_INET */
 201 static inline
 202 int lttng_enumerate_network_ip_interface(struct lttng_session *session)
 203 {
 204         return 0;
 205 }
 206 #endif /* CONFIG_INET */
 207
 208 static
 209 int lttng_dump_one_fd(const void *p, struct file *file, unsigned int fd)
 210 {
 211         const struct lttng_fd_ctx *ctx = p;
 212         const char *s = d_path(&file->f_path, ctx->page, PAGE_SIZE);
 213         unsigned int flags = file->f_flags;
 214
 215         /*
 216          * We don't expose kernel internal flags, only userspace-visible
 217          * flags.
 218          */
 219         flags &= ~FMODE_NONOTIFY;
 220         if (test_bit(fd, ctx->fdt->close_on_exec))
 221                 flags |= O_CLOEXEC;
 222         if (IS_ERR(s)) {
 223                 struct dentry *dentry = file->f_path.dentry;
 224
 225                 /* Make sure we give at least some info */
 226                 spin_lock(&dentry->d_lock);
 227                 trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd,
 228                         dentry->d_name.name, flags, file->f_mode);
 229                 spin_unlock(&dentry->d_lock);
 230                 goto end;
 231         }
 232         trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd, s,
 233                 flags, file->f_mode);
 234 end:
 235         return 0;
 236 }
 237
 238 static
 239 void lttng_enumerate_task_fd(struct lttng_session *session,
 240                 struct task_struct *p, char *tmp)
 241 {
 242         struct lttng_fd_ctx ctx = { .page = tmp, .session = session, .p = p };
 243
 244         task_lock(p);
 245         ctx.fdt = files_fdtable(p->files);
 246         lttng_iterate_fd(p->files, 0, lttng_dump_one_fd, &ctx);
 247         task_unlock(p);
 248 }
 249
 250 static
 251 int lttng_enumerate_file_descriptors(struct lttng_session *session)
 252 {
 253         struct task_struct *p;
 254         char *tmp = (char *) __get_free_page(GFP_KERNEL);
 255
 256         /* Enumerate active file descriptors */
 257         rcu_read_lock();
 258         for_each_process(p)
 259                 lttng_enumerate_task_fd(session, p, tmp);
 260         rcu_read_unlock();
 261         free_page((unsigned long) tmp);
 262         return 0;
 263 }
 264
 265 #if 0
 266 /*
 267  * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
 268  * (scheduling in atomic). Normally, the tasklist lock protects this kind of
 269  * iteration, but it is not exported to modules.
 270  */
 271 static
 272 void lttng_enumerate_task_vm_maps(struct lttng_session *session,
 273                 struct task_struct *p)
 274 {
 275         struct mm_struct *mm;
 276         struct vm_area_struct *map;
 277         unsigned long ino;
 278
 279         /* get_task_mm does a task_lock... */
 280         mm = get_task_mm(p);
 281         if (!mm)
 282                 return;
 283
 284         map = mm->mmap;
 285         if (map) {
 286                 down_read(&mm->mmap_sem);
 287                 while (map) {
 288                         if (map->vm_file)
 289                                 ino = map->vm_file->f_dentry->d_inode->i_ino;
 290                         else
 291                                 ino = 0;
 292                         trace_lttng_statedump_vm_map(session, p, map, ino);
 293                         map = map->vm_next;
 294                 }
 295                 up_read(&mm->mmap_sem);
 296         }
 297         mmput(mm);
 298 }
 299
 300 static
 301 int lttng_enumerate_vm_maps(struct lttng_session *session)
 302 {
 303         struct task_struct *p;
 304
 305         rcu_read_lock();
 306         for_each_process(p)
 307                 lttng_enumerate_task_vm_maps(session, p);
 308         rcu_read_unlock();
 309         return 0;
 310 }
 311 #endif
 312
 313 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
 314
 315 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39))
 316 #define irq_desc_get_chip(desc) get_irq_desc_chip(desc)
 317 #endif
 318
 319 static
 320 void lttng_list_interrupts(struct lttng_session *session)
 321 {
 322         unsigned int irq;
 323         unsigned long flags = 0;
 324         struct irq_desc *desc;
 325
 326 #define irq_to_desc     wrapper_irq_to_desc
 327         /* needs irq_desc */
 328         for_each_irq_desc(irq, desc) {
 329                 struct irqaction *action;
 330                 const char *irq_chip_name =
 331                         irq_desc_get_chip(desc)->name ? : "unnamed_irq_chip";
 332
 333                 local_irq_save(flags);
 334                 wrapper_desc_spin_lock(&desc->lock);
 335                 for (action = desc->action; action; action = action->next) {
 336                         trace_lttng_statedump_interrupt(session,
 337                                 irq, irq_chip_name, action);
 338                 }
 339                 wrapper_desc_spin_unlock(&desc->lock);
 340                 local_irq_restore(flags);
 341         }
 342 #undef irq_to_desc
 343 }
 344 #else
 345 static inline
 346 void lttng_list_interrupts(struct lttng_session *session)
 347 {
 348 }
 349 #endif
 350
 351 static
 352 void lttng_statedump_process_ns(struct lttng_session *session,
 353                 struct task_struct *p,
 354                 enum lttng_thread_type type,
 355                 enum lttng_execution_mode mode,
 356                 enum lttng_execution_submode submode,
 357                 enum lttng_process_status status)
 358 {
 359         struct nsproxy *proxy;
 360         struct pid_namespace *pid_ns;
 361
 362         rcu_read_lock();
 363         proxy = task_nsproxy(p);
 364         if (proxy) {
 365                 pid_ns = lttng_get_proxy_pid_ns(proxy);
 366                 do {
 367                         trace_lttng_statedump_process_state(session,
 368                                 p, type, mode, submode, status, pid_ns);
 369                         pid_ns = pid_ns->parent;
 370                 } while (pid_ns);
 371         } else {
 372                 trace_lttng_statedump_process_state(session,
 373                         p, type, mode, submode, status, NULL);
 374         }
 375         rcu_read_unlock();
 376 }
 377
 378 static
 379 int lttng_enumerate_process_states(struct lttng_session *session)
 380 {
 381         struct task_struct *g, *p;
 382
 383         rcu_read_lock();
 384         for_each_process(g) {
 385                 p = g;
 386                 do {
 387                         enum lttng_execution_mode mode =
 388                                 LTTNG_MODE_UNKNOWN;
 389                         enum lttng_execution_submode submode =
 390                                 LTTNG_UNKNOWN;
 391                         enum lttng_process_status status;
 392                         enum lttng_thread_type type;
 393
 394                         task_lock(p);
 395                         if (p->exit_state == EXIT_ZOMBIE)
 396                                 status = LTTNG_ZOMBIE;
 397                         else if (p->exit_state == EXIT_DEAD)
 398                                 status = LTTNG_DEAD;
 399                         else if (p->state == TASK_RUNNING) {
 400                                 /* Is this a forked child that has not run yet? */
 401                                 if (list_empty(&p->rt.run_list))
 402                                         status = LTTNG_WAIT_FORK;
 403                                 else
 404                                         /*
 405                                          * All tasks are considered as wait_cpu;
 406                                          * the viewer will sort out if the task
 407                                          * was really running at this time.
 408                                          */
 409                                         status = LTTNG_WAIT_CPU;
 410                         } else if (p->state &
 411                                 (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)) {
 412                                 /* Task is waiting for something to complete */
 413                                 status = LTTNG_WAIT;
 414                         } else
 415                                 status = LTTNG_UNNAMED;
 416                         submode = LTTNG_NONE;
 417
 418                         /*
 419                          * Verification of t->mm is to filter out kernel
 420                          * threads; Viewer will further filter out if a
 421                          * user-space thread was in syscall mode or not.
 422                          */
 423                         if (p->mm)
 424                                 type = LTTNG_USER_THREAD;
 425                         else
 426                                 type = LTTNG_KERNEL_THREAD;
 427                         lttng_statedump_process_ns(session,
 428                                 p, type, mode, submode, status);
 429                         task_unlock(p);
 430                 } while_each_thread(g, p);
 431         }
 432         rcu_read_unlock();
 433
 434         return 0;
 435 }
 436
 437 static
 438 void lttng_statedump_work_func(struct work_struct *work)
 439 {
 440         if (atomic_dec_and_test(&kernel_threads_to_run))
 441                 /* If we are the last thread, wake up do_lttng_statedump */
 442                 wake_up(&statedump_wq);
 443 }
 444
 445 static
 446 int do_lttng_statedump(struct lttng_session *session)
 447 {
 448         int cpu;
 449
 450         trace_lttng_statedump_start(session);
 451         lttng_enumerate_process_states(session);
 452         lttng_enumerate_file_descriptors(session);
 453         /* FIXME lttng_enumerate_vm_maps(session); */
 454         lttng_list_interrupts(session);
 455         lttng_enumerate_network_ip_interface(session);
 456         lttng_enumerate_block_devices(session);
 457
 458         /* TODO lttng_dump_idt_table(session); */
 459         /* TODO lttng_dump_softirq_vec(session); */
 460         /* TODO lttng_list_modules(session); */
 461         /* TODO lttng_dump_swap_files(session); */
 462
 463         /*
 464          * Fire off a work queue on each CPU. Their sole purpose in life
 465          * is to guarantee that each CPU has been in a state where is was in
 466          * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
 467          */
 468         get_online_cpus();
 469         atomic_set(&kernel_threads_to_run, num_online_cpus());
 470         for_each_online_cpu(cpu) {
 471                 INIT_DELAYED_WORK(&cpu_work[cpu], lttng_statedump_work_func);
 472                 schedule_delayed_work_on(cpu, &cpu_work[cpu], 0);
 473         }
 474         /* Wait for all threads to run */
 475         __wait_event(statedump_wq, (atomic_read(&kernel_threads_to_run) == 0));
 476         put_online_cpus();
 477         /* Our work is done */
 478         trace_lttng_statedump_end(session);
 479         return 0;
 480 }
 481
 482 /*
 483  * Called with session mutex held.
 484  */
 485 int lttng_statedump_start(struct lttng_session *session)
 486 {
 487         return do_lttng_statedump(session);
 488 }
 489 EXPORT_SYMBOL_GPL(lttng_statedump_start);
 490
 491 static
 492 int __init lttng_statedump_init(void)
 493 {
 494         /*
 495          * Allow module to load even if the fixup cannot be done. This
 496          * will allow seemless transition when the underlying issue fix
 497          * is merged into the Linux kernel, and when tracepoint.c
 498          * "tracepoint_module_notify" is turned into a static function.
 499          */
 500         (void) wrapper_lttng_fixup_sig(THIS_MODULE);
 501         return 0;
 502 }
 503
 504 module_init(lttng_statedump_init);
 505
 506 static
 507 void __exit lttng_statedump_exit(void)
 508 {
 509 }
 510
 511 module_exit(lttng_statedump_exit);
 512
 513 MODULE_LICENSE("GPL and additional rights");
 514 MODULE_AUTHOR("Jean-Hugues Deschenes");
 515 MODULE_DESCRIPTION("Linux Trace Toolkit Next Generation Statedump");
 516 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION) "."
 517         __stringify(LTTNG_MODULES_MINOR_VERSION) "."
 518         __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION)
 519         LTTNG_MODULES_EXTRAVERSION);