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 };
  83
  84 /*
  85  * Protected by the trace lock.
  86  */
  87 static struct delayed_work cpu_work[NR_CPUS];
  88 static DECLARE_WAIT_QUEUE_HEAD(statedump_wq);
  89 static atomic_t kernel_threads_to_run;
  90
  91 enum lttng_thread_type {
  92         LTTNG_USER_THREAD = 0,
  93         LTTNG_KERNEL_THREAD = 1,
  94 };
  95
  96 enum lttng_execution_mode {
  97         LTTNG_USER_MODE = 0,
  98         LTTNG_SYSCALL = 1,
  99         LTTNG_TRAP = 2,
 100         LTTNG_IRQ = 3,
 101         LTTNG_SOFTIRQ = 4,
 102         LTTNG_MODE_UNKNOWN = 5,
 103 };
 104
 105 enum lttng_execution_submode {
 106         LTTNG_NONE = 0,
 107         LTTNG_UNKNOWN = 1,
 108 };
 109
 110 enum lttng_process_status {
 111         LTTNG_UNNAMED = 0,
 112         LTTNG_WAIT_FORK = 1,
 113         LTTNG_WAIT_CPU = 2,
 114         LTTNG_EXIT = 3,
 115         LTTNG_ZOMBIE = 4,
 116         LTTNG_WAIT = 5,
 117         LTTNG_RUN = 6,
 118         LTTNG_DEAD = 7,
 119 };
 120
 121 static
 122 int lttng_enumerate_block_devices(struct lttng_session *session)
 123 {
 124         struct class *ptr_block_class;
 125         struct device_type *ptr_disk_type;
 126         struct class_dev_iter iter;
 127         struct device *dev;
 128
 129         ptr_block_class = wrapper_get_block_class();
 130         if (!ptr_block_class)
 131                 return -ENOSYS;
 132         ptr_disk_type = wrapper_get_disk_type();
 133         if (!ptr_disk_type) {
 134                 return -ENOSYS;
 135         }
 136         class_dev_iter_init(&iter, ptr_block_class, NULL, ptr_disk_type);
 137         while ((dev = class_dev_iter_next(&iter))) {
 138                 struct disk_part_iter piter;
 139                 struct gendisk *disk = dev_to_disk(dev);
 140                 struct hd_struct *part;
 141
 142                 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
 143                 while ((part = disk_part_iter_next(&piter))) {
 144                         char name_buf[BDEVNAME_SIZE];
 145                         char *p;
 146
 147                         p = wrapper_disk_name(disk, part->partno, name_buf);
 148                         if (!p) {
 149                                 disk_part_iter_exit(&piter);
 150                                 class_dev_iter_exit(&iter);
 151                                 return -ENOSYS;
 152                         }
 153                         trace_lttng_statedump_block_device(session,
 154                                         part_devt(part), name_buf);
 155                 }
 156                 disk_part_iter_exit(&piter);
 157         }
 158         class_dev_iter_exit(&iter);
 159         return 0;
 160 }
 161
 162 #ifdef CONFIG_INET
 163
 164 static
 165 void lttng_enumerate_device(struct lttng_session *session,
 166                 struct net_device *dev)
 167 {
 168         struct in_device *in_dev;
 169         struct in_ifaddr *ifa;
 170
 171         if (dev->flags & IFF_UP) {
 172                 in_dev = in_dev_get(dev);
 173                 if (in_dev) {
 174                         for (ifa = in_dev->ifa_list; ifa != NULL;
 175                              ifa = ifa->ifa_next) {
 176                                 trace_lttng_statedump_network_interface(
 177                                         session, dev, ifa);
 178                         }
 179                         in_dev_put(in_dev);
 180                 }
 181         } else {
 182                 trace_lttng_statedump_network_interface(
 183                         session, dev, NULL);
 184         }
 185 }
 186
 187 static
 188 int lttng_enumerate_network_ip_interface(struct lttng_session *session)
 189 {
 190         struct net_device *dev;
 191
 192         read_lock(&dev_base_lock);
 193         for_each_netdev(&init_net, dev)
 194                 lttng_enumerate_device(session, dev);
 195         read_unlock(&dev_base_lock);
 196
 197         return 0;
 198 }
 199 #else /* CONFIG_INET */
 200 static inline
 201 int lttng_enumerate_network_ip_interface(struct lttng_session *session)
 202 {
 203         return 0;
 204 }
 205 #endif /* CONFIG_INET */
 206
 207 static
 208 int lttng_dump_one_fd(const void *p, struct file *file, unsigned int fd)
 209 {
 210         const struct lttng_fd_ctx *ctx = p;
 211         const char *s = d_path(&file->f_path, ctx->page, PAGE_SIZE);
 212
 213         if (IS_ERR(s)) {
 214                 struct dentry *dentry = file->f_path.dentry;
 215
 216                 /* Make sure we give at least some info */
 217                 spin_lock(&dentry->d_lock);
 218                 trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd,
 219                         dentry->d_name.name);
 220                 spin_unlock(&dentry->d_lock);
 221                 goto end;
 222         }
 223         trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd, s);
 224 end:
 225         return 0;
 226 }
 227
 228 static
 229 void lttng_enumerate_task_fd(struct lttng_session *session,
 230                 struct task_struct *p, char *tmp)
 231 {
 232         struct lttng_fd_ctx ctx = { .page = tmp, .session = session, .p = p };
 233
 234         task_lock(p);
 235         lttng_iterate_fd(p->files, 0, lttng_dump_one_fd, &ctx);
 236         task_unlock(p);
 237 }
 238
 239 static
 240 int lttng_enumerate_file_descriptors(struct lttng_session *session)
 241 {
 242         struct task_struct *p;
 243         char *tmp = (char *) __get_free_page(GFP_KERNEL);
 244
 245         /* Enumerate active file descriptors */
 246         rcu_read_lock();
 247         for_each_process(p)
 248                 lttng_enumerate_task_fd(session, p, tmp);
 249         rcu_read_unlock();
 250         free_page((unsigned long) tmp);
 251         return 0;
 252 }
 253
 254 #if 0
 255 /*
 256  * FIXME: we cannot take a mmap_sem while in a RCU read-side critical section
 257  * (scheduling in atomic). Normally, the tasklist lock protects this kind of
 258  * iteration, but it is not exported to modules.
 259  */
 260 static
 261 void lttng_enumerate_task_vm_maps(struct lttng_session *session,
 262                 struct task_struct *p)
 263 {
 264         struct mm_struct *mm;
 265         struct vm_area_struct *map;
 266         unsigned long ino;
 267
 268         /* get_task_mm does a task_lock... */
 269         mm = get_task_mm(p);
 270         if (!mm)
 271                 return;
 272
 273         map = mm->mmap;
 274         if (map) {
 275                 down_read(&mm->mmap_sem);
 276                 while (map) {
 277                         if (map->vm_file)
 278                                 ino = map->vm_file->f_dentry->d_inode->i_ino;
 279                         else
 280                                 ino = 0;
 281                         trace_lttng_statedump_vm_map(session, p, map, ino);
 282                         map = map->vm_next;
 283                 }
 284                 up_read(&mm->mmap_sem);
 285         }
 286         mmput(mm);
 287 }
 288
 289 static
 290 int lttng_enumerate_vm_maps(struct lttng_session *session)
 291 {
 292         struct task_struct *p;
 293
 294         rcu_read_lock();
 295         for_each_process(p)
 296                 lttng_enumerate_task_vm_maps(session, p);
 297         rcu_read_unlock();
 298         return 0;
 299 }
 300 #endif
 301
 302 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
 303
 304 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39))
 305 #define irq_desc_get_chip(desc) get_irq_desc_chip(desc)
 306 #endif
 307
 308 static
 309 void lttng_list_interrupts(struct lttng_session *session)
 310 {
 311         unsigned int irq;
 312         unsigned long flags = 0;
 313         struct irq_desc *desc;
 314
 315 #define irq_to_desc     wrapper_irq_to_desc
 316         /* needs irq_desc */
 317         for_each_irq_desc(irq, desc) {
 318                 struct irqaction *action;
 319                 const char *irq_chip_name =
 320                         irq_desc_get_chip(desc)->name ? : "unnamed_irq_chip";
 321
 322                 local_irq_save(flags);
 323                 wrapper_desc_spin_lock(&desc->lock);
 324                 for (action = desc->action; action; action = action->next) {
 325                         trace_lttng_statedump_interrupt(session,
 326                                 irq, irq_chip_name, action);
 327                 }
 328                 wrapper_desc_spin_unlock(&desc->lock);
 329                 local_irq_restore(flags);
 330         }
 331 #undef irq_to_desc
 332 }
 333 #else
 334 static inline
 335 void lttng_list_interrupts(struct lttng_session *session)
 336 {
 337 }
 338 #endif
 339
 340 static
 341 void lttng_statedump_process_ns(struct lttng_session *session,
 342                 struct task_struct *p,
 343                 enum lttng_thread_type type,
 344                 enum lttng_execution_mode mode,
 345                 enum lttng_execution_submode submode,
 346                 enum lttng_process_status status)
 347 {
 348         struct nsproxy *proxy;
 349         struct pid_namespace *pid_ns;
 350
 351         rcu_read_lock();
 352         proxy = task_nsproxy(p);
 353         if (proxy) {
 354                 pid_ns = lttng_get_proxy_pid_ns(proxy);
 355                 do {
 356                         trace_lttng_statedump_process_state(session,
 357                                 p, type, mode, submode, status, pid_ns);
 358                         pid_ns = pid_ns->parent;
 359                 } while (pid_ns);
 360         } else {
 361                 trace_lttng_statedump_process_state(session,
 362                         p, type, mode, submode, status, NULL);
 363         }
 364         rcu_read_unlock();
 365 }
 366
 367 static
 368 int lttng_enumerate_process_states(struct lttng_session *session)
 369 {
 370         struct task_struct *g, *p;
 371
 372         rcu_read_lock();
 373         for_each_process(g) {
 374                 p = g;
 375                 do {
 376                         enum lttng_execution_mode mode =
 377                                 LTTNG_MODE_UNKNOWN;
 378                         enum lttng_execution_submode submode =
 379                                 LTTNG_UNKNOWN;
 380                         enum lttng_process_status status;
 381                         enum lttng_thread_type type;
 382
 383                         task_lock(p);
 384                         if (p->exit_state == EXIT_ZOMBIE)
 385                                 status = LTTNG_ZOMBIE;
 386                         else if (p->exit_state == EXIT_DEAD)
 387                                 status = LTTNG_DEAD;
 388                         else if (p->state == TASK_RUNNING) {
 389                                 /* Is this a forked child that has not run yet? */
 390                                 if (list_empty(&p->rt.run_list))
 391                                         status = LTTNG_WAIT_FORK;
 392                                 else
 393                                         /*
 394                                          * All tasks are considered as wait_cpu;
 395                                          * the viewer will sort out if the task
 396                                          * was really running at this time.
 397                                          */
 398                                         status = LTTNG_WAIT_CPU;
 399                         } else if (p->state &
 400                                 (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)) {
 401                                 /* Task is waiting for something to complete */
 402                                 status = LTTNG_WAIT;
 403                         } else
 404                                 status = LTTNG_UNNAMED;
 405                         submode = LTTNG_NONE;
 406
 407                         /*
 408                          * Verification of t->mm is to filter out kernel
 409                          * threads; Viewer will further filter out if a
 410                          * user-space thread was in syscall mode or not.
 411                          */
 412                         if (p->mm)
 413                                 type = LTTNG_USER_THREAD;
 414                         else
 415                                 type = LTTNG_KERNEL_THREAD;
 416                         lttng_statedump_process_ns(session,
 417                                 p, type, mode, submode, status);
 418                         task_unlock(p);
 419                 } while_each_thread(g, p);
 420         }
 421         rcu_read_unlock();
 422
 423         return 0;
 424 }
 425
 426 static
 427 void lttng_statedump_work_func(struct work_struct *work)
 428 {
 429         if (atomic_dec_and_test(&kernel_threads_to_run))
 430                 /* If we are the last thread, wake up do_lttng_statedump */
 431                 wake_up(&statedump_wq);
 432 }
 433
 434 static
 435 int do_lttng_statedump(struct lttng_session *session)
 436 {
 437         int cpu;
 438
 439         trace_lttng_statedump_start(session);
 440         lttng_enumerate_process_states(session);
 441         lttng_enumerate_file_descriptors(session);
 442         /* FIXME lttng_enumerate_vm_maps(session); */
 443         lttng_list_interrupts(session);
 444         lttng_enumerate_network_ip_interface(session);
 445         lttng_enumerate_block_devices(session);
 446
 447         /* TODO lttng_dump_idt_table(session); */
 448         /* TODO lttng_dump_softirq_vec(session); */
 449         /* TODO lttng_list_modules(session); */
 450         /* TODO lttng_dump_swap_files(session); */
 451
 452         /*
 453          * Fire off a work queue on each CPU. Their sole purpose in life
 454          * is to guarantee that each CPU has been in a state where is was in
 455          * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
 456          */
 457         get_online_cpus();
 458         atomic_set(&kernel_threads_to_run, num_online_cpus());
 459         for_each_online_cpu(cpu) {
 460                 INIT_DELAYED_WORK(&cpu_work[cpu], lttng_statedump_work_func);
 461                 schedule_delayed_work_on(cpu, &cpu_work[cpu], 0);
 462         }
 463         /* Wait for all threads to run */
 464         __wait_event(statedump_wq, (atomic_read(&kernel_threads_to_run) == 0));
 465         put_online_cpus();
 466         /* Our work is done */
 467         trace_lttng_statedump_end(session);
 468         return 0;
 469 }
 470
 471 /*
 472  * Called with session mutex held.
 473  */
 474 int lttng_statedump_start(struct lttng_session *session)
 475 {
 476         return do_lttng_statedump(session);
 477 }
 478 EXPORT_SYMBOL_GPL(lttng_statedump_start);
 479
 480 static
 481 int __init lttng_statedump_init(void)
 482 {
 483         /*
 484          * Allow module to load even if the fixup cannot be done. This
 485          * will allow seemless transition when the underlying issue fix
 486          * is merged into the Linux kernel, and when tracepoint.c
 487          * "tracepoint_module_notify" is turned into a static function.
 488          */
 489         (void) wrapper_lttng_fixup_sig(THIS_MODULE);
 490         return 0;
 491 }
 492
 493 module_init(lttng_statedump_init);
 494
 495 static
 496 void __exit lttng_statedump_exit(void)
 497 {
 498 }
 499
 500 module_exit(lttng_statedump_exit);
 501
 502 MODULE_LICENSE("GPL and additional rights");
 503 MODULE_AUTHOR("Jean-Hugues Deschenes");
 504 MODULE_DESCRIPTION("Linux Trace Toolkit Next Generation Statedump");
 505 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION) "."
 506         __stringify(LTTNG_MODULES_MINOR_VERSION) "."
 507         __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION)
 508         LTTNG_MODULES_EXTRAVERSION);