kernel/marker.c

   1 /*
   2  * Copyright (C) 2007 Mathieu Desnoyers
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License as published by
   6  * the Free Software Foundation; either version 2 of the License, or
   7  * (at your option) any later version.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write to the Free Software
  16  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  17  */
  18 #include <linux/module.h>
  19 #include <linux/mutex.h>
  20 #include <linux/types.h>
  21 #include <linux/jhash.h>
  22 #include <linux/list.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/marker.h>
  25 #include <linux/err.h>
  26 #include <linux/slab.h>
  27
  28 extern struct marker __start___markers[];
  29 extern struct marker __stop___markers[];
  30
  31 /* Set to 1 to enable marker debug output */
  32 static const int marker_debug;
  33
  34 /*
  35  * markers_mutex nests inside module_mutex. Markers mutex protects the builtin
  36  * and module markers and the hash table.
  37  */
  38 static DEFINE_MUTEX(markers_mutex);
  39
  40 /*
  41  * Marker hash table, containing the active markers.
  42  * Protected by module_mutex.
  43  */
  44 #define MARKER_HASH_BITS 6
  45 #define MARKER_TABLE_SIZE (1 << MARKER_HASH_BITS)
  46 static struct hlist_head marker_table[MARKER_TABLE_SIZE];
  47
  48 /*
  49  * Note about RCU :
  50  * It is used to make sure every handler has finished using its private data
  51  * between two consecutive operation (add or remove) on a given marker.  It is
  52  * also used to delay the free of multiple probes array until a quiescent state
  53  * is reached.
  54  * marker entries modifications are protected by the markers_mutex.
  55  */
  56 struct marker_entry {
  57         struct hlist_node hlist;
  58         char *format;
  59                         /* Probe wrapper */
  60         void (*call)(const struct marker *mdata, void *call_private, ...);
  61         struct marker_probe_closure single;
  62         struct marker_probe_closure *multi;
  63         int refcount;   /* Number of times armed. 0 if disarmed. */
  64         struct rcu_head rcu;
  65         void *oldptr;
  66         int rcu_pending;
  67         unsigned char ptype:1;
  68         unsigned char format_allocated:1;
  69         char name[0];   /* Contains name'\0'format'\0' */
  70 };
  71
  72 /**
  73  * __mark_empty_function - Empty probe callback
  74  * @probe_private: probe private data
  75  * @call_private: call site private data
  76  * @fmt: format string
  77  * @...: variable argument list
  78  *
  79  * Empty callback provided as a probe to the markers. By providing this to a
  80  * disabled marker, we make sure the  execution flow is always valid even
  81  * though the function pointer change and the marker enabling are two distinct
  82  * operations that modifies the execution flow of preemptible code.
  83  */
  84 notrace void __mark_empty_function(void *probe_private, void *call_private,
  85         const char *fmt, va_list *args)
  86 {
  87 }
  88 EXPORT_SYMBOL_GPL(__mark_empty_function);
  89
  90 /*
  91  * marker_probe_cb Callback that prepares the variable argument list for probes.
  92  * @mdata: pointer of type struct marker
  93  * @call_private: caller site private data
  94  * @...:  Variable argument list.
  95  *
  96  * Since we do not use "typical" pointer based RCU in the 1 argument case, we
  97  * need to put a full smp_rmb() in this branch. This is why we do not use
  98  * rcu_dereference() for the pointer read.
  99  */
 100 notrace void marker_probe_cb(const struct marker *mdata,
 101                 void *call_private, ...)
 102 {
 103         va_list args;
 104         char ptype;
 105
 106         /*
 107          * rcu_read_lock_sched does two things : disabling preemption to make
 108          * sure the teardown of the callbacks can be done correctly when they
 109          * are in modules and they insure RCU read coherency.
 110          */
 111         rcu_read_lock_sched_notrace();
 112         ptype = mdata->ptype;
 113         if (likely(!ptype)) {
 114                 marker_probe_func *func;
 115                 /* Must read the ptype before ptr. They are not data dependant,
 116                  * so we put an explicit smp_rmb() here. */
 117                 smp_rmb();
 118                 func = mdata->single.func;
 119                 /* Must read the ptr before private data. They are not data
 120                  * dependant, so we put an explicit smp_rmb() here. */
 121                 smp_rmb();
 122                 va_start(args, call_private);
 123                 func(mdata->single.probe_private, call_private, mdata->format,
 124                         &args);
 125                 va_end(args);
 126         } else {
 127                 struct marker_probe_closure *multi;
 128                 int i;
 129                 /*
 130                  * Read mdata->ptype before mdata->multi.
 131                  */
 132                 smp_rmb();
 133                 multi = mdata->multi;
 134                 /*
 135                  * multi points to an array, therefore accessing the array
 136                  * depends on reading multi. However, even in this case,
 137                  * we must insure that the pointer is read _before_ the array
 138                  * data. Same as rcu_dereference, but we need a full smp_rmb()
 139                  * in the fast path, so put the explicit barrier here.
 140                  */
 141                 smp_read_barrier_depends();
 142                 for (i = 0; multi[i].func; i++) {
 143                         va_start(args, call_private);
 144                         multi[i].func(multi[i].probe_private, call_private,
 145                                 mdata->format, &args);
 146                         va_end(args);
 147                 }
 148         }
 149         rcu_read_unlock_sched_notrace();
 150 }
 151 EXPORT_SYMBOL_GPL(marker_probe_cb);
 152
 153 /*
 154  * marker_probe_cb Callback that does not prepare the variable argument list.
 155  * @mdata: pointer of type struct marker
 156  * @call_private: caller site private data
 157  * @...:  Variable argument list.
 158  *
 159  * Should be connected to markers "MARK_NOARGS".
 160  */
 161 static notrace void marker_probe_cb_noarg(const struct marker *mdata,
 162                 void *call_private, ...)
 163 {
 164         va_list args;   /* not initialized */
 165         char ptype;
 166
 167         rcu_read_lock_sched_notrace();
 168         ptype = mdata->ptype;
 169         if (likely(!ptype)) {
 170                 marker_probe_func *func;
 171                 /* Must read the ptype before ptr. They are not data dependant,
 172                  * so we put an explicit smp_rmb() here. */
 173                 smp_rmb();
 174                 func = mdata->single.func;
 175                 /* Must read the ptr before private data. They are not data
 176                  * dependant, so we put an explicit smp_rmb() here. */
 177                 smp_rmb();
 178                 func(mdata->single.probe_private, call_private, mdata->format,
 179                         &args);
 180         } else {
 181                 struct marker_probe_closure *multi;
 182                 int i;
 183                 /*
 184                  * Read mdata->ptype before mdata->multi.
 185                  */
 186                 smp_rmb();
 187                 multi = mdata->multi;
 188                 /*
 189                  * multi points to an array, therefore accessing the array
 190                  * depends on reading multi. However, even in this case,
 191                  * we must insure that the pointer is read _before_ the array
 192                  * data. Same as rcu_dereference, but we need a full smp_rmb()
 193                  * in the fast path, so put the explicit barrier here.
 194                  */
 195                 smp_read_barrier_depends();
 196                 for (i = 0; multi[i].func; i++)
 197                         multi[i].func(multi[i].probe_private, call_private,
 198                                 mdata->format, &args);
 199         }
 200         rcu_read_unlock_sched_notrace();
 201 }
 202
 203 static void free_old_closure(struct rcu_head *head)
 204 {
 205         struct marker_entry *entry = container_of(head,
 206                 struct marker_entry, rcu);
 207         kfree(entry->oldptr);
 208         /* Make sure we free the data before setting the pending flag to 0 */
 209         smp_wmb();
 210         entry->rcu_pending = 0;
 211 }
 212
 213 static void debug_print_probes(struct marker_entry *entry)
 214 {
 215         int i;
 216
 217         if (!marker_debug)
 218                 return;
 219
 220         if (!entry->ptype) {
 221                 printk(KERN_DEBUG "Single probe : %p %p\n",
 222                         entry->single.func,
 223                         entry->single.probe_private);
 224         } else {
 225                 for (i = 0; entry->multi[i].func; i++)
 226                         printk(KERN_DEBUG "Multi probe %d : %p %p\n", i,
 227                                 entry->multi[i].func,
 228                                 entry->multi[i].probe_private);
 229         }
 230 }
 231
 232 static struct marker_probe_closure *
 233 marker_entry_add_probe(struct marker_entry *entry,
 234                 marker_probe_func *probe, void *probe_private)
 235 {
 236         int nr_probes = 0;
 237         struct marker_probe_closure *old, *new;
 238
 239         WARN_ON(!probe);
 240
 241         debug_print_probes(entry);
 242         old = entry->multi;
 243         if (!entry->ptype) {
 244                 if (entry->single.func == probe &&
 245                                 entry->single.probe_private == probe_private)
 246                         return ERR_PTR(-EBUSY);
 247                 if (entry->single.func == __mark_empty_function) {
 248                         /* 0 -> 1 probes */
 249                         entry->single.func = probe;
 250                         entry->single.probe_private = probe_private;
 251                         entry->refcount = 1;
 252                         entry->ptype = 0;
 253                         debug_print_probes(entry);
 254                         return NULL;
 255                 } else {
 256                         /* 1 -> 2 probes */
 257                         nr_probes = 1;
 258                         old = NULL;
 259                 }
 260         } else {
 261                 /* (N -> N+1), (N != 0, 1) probes */
 262                 for (nr_probes = 0; old[nr_probes].func; nr_probes++)
 263                         if (old[nr_probes].func == probe
 264                                         && old[nr_probes].probe_private
 265                                                 == probe_private)
 266                                 return ERR_PTR(-EBUSY);
 267         }
 268         /* + 2 : one for new probe, one for NULL func */
 269         new = kzalloc((nr_probes + 2) * sizeof(struct marker_probe_closure),
 270                         GFP_KERNEL);
 271         if (new == NULL)
 272                 return ERR_PTR(-ENOMEM);
 273         if (!old)
 274                 new[0] = entry->single;
 275         else
 276                 memcpy(new, old,
 277                         nr_probes * sizeof(struct marker_probe_closure));
 278         new[nr_probes].func = probe;
 279         new[nr_probes].probe_private = probe_private;
 280         entry->refcount = nr_probes + 1;
 281         entry->multi = new;
 282         entry->ptype = 1;
 283         debug_print_probes(entry);
 284         return old;
 285 }
 286
 287 static struct marker_probe_closure *
 288 marker_entry_remove_probe(struct marker_entry *entry,
 289                 marker_probe_func *probe, void *probe_private)
 290 {
 291         int nr_probes = 0, nr_del = 0, i;
 292         struct marker_probe_closure *old, *new;
 293
 294         old = entry->multi;
 295
 296         debug_print_probes(entry);
 297         if (!entry->ptype) {
 298                 /* 0 -> N is an error */
 299                 WARN_ON(entry->single.func == __mark_empty_function);
 300                 /* 1 -> 0 probes */
 301                 WARN_ON(probe && entry->single.func != probe);
 302                 WARN_ON(entry->single.probe_private != probe_private);
 303                 entry->single.func = __mark_empty_function;
 304                 entry->refcount = 0;
 305                 entry->ptype = 0;
 306                 debug_print_probes(entry);
 307                 return NULL;
 308         } else {
 309                 /* (N -> M), (N > 1, M >= 0) probes */
 310                 for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
 311                         if ((!probe || old[nr_probes].func == probe)
 312                                         && old[nr_probes].probe_private
 313                                                 == probe_private)
 314                                 nr_del++;
 315                 }
 316         }
 317
 318         if (nr_probes - nr_del == 0) {
 319                 /* N -> 0, (N > 1) */
 320                 entry->single.func = __mark_empty_function;
 321                 entry->refcount = 0;
 322                 entry->ptype = 0;
 323         } else if (nr_probes - nr_del == 1) {
 324                 /* N -> 1, (N > 1) */
 325                 for (i = 0; old[i].func; i++)
 326                         if ((probe && old[i].func != probe) ||
 327                                         old[i].probe_private != probe_private)
 328                                 entry->single = old[i];
 329                 entry->refcount = 1;
 330                 entry->ptype = 0;
 331         } else {
 332                 int j = 0;
 333                 /* N -> M, (N > 1, M > 1) */
 334                 /* + 1 for NULL */
 335                 new = kzalloc((nr_probes - nr_del + 1)
 336                         * sizeof(struct marker_probe_closure), GFP_KERNEL);
 337                 if (new == NULL)
 338                         return ERR_PTR(-ENOMEM);
 339                 for (i = 0; old[i].func; i++)
 340                         if ((probe && old[i].func != probe) ||
 341                                         old[i].probe_private != probe_private)
 342                                 new[j++] = old[i];
 343                 entry->refcount = nr_probes - nr_del;
 344                 entry->ptype = 1;
 345                 entry->multi = new;
 346         }
 347         debug_print_probes(entry);
 348         return old;
 349 }
 350
 351 /*
 352  * Get marker if the marker is present in the marker hash table.
 353  * Must be called with markers_mutex held.
 354  * Returns NULL if not present.
 355  */
 356 static struct marker_entry *get_marker(const char *name)
 357 {
 358         struct hlist_head *head;
 359         struct hlist_node *node;
 360         struct marker_entry *e;
 361         u32 hash = jhash(name, strlen(name), 0);
 362
 363         head = &marker_table[hash & ((1 << MARKER_HASH_BITS)-1)];
 364         hlist_for_each_entry(e, node, head, hlist) {
 365                 if (!strcmp(name, e->name))
 366                         return e;
 367         }
 368         return NULL;
 369 }
 370
 371 /*
 372  * Add the marker to the marker hash table. Must be called with markers_mutex
 373  * held.
 374  */
 375 static struct marker_entry *add_marker(const char *name, const char *format)
 376 {
 377         struct hlist_head *head;
 378         struct hlist_node *node;
 379         struct marker_entry *e;
 380         size_t name_len = strlen(name) + 1;
 381         size_t format_len = 0;
 382         u32 hash = jhash(name, name_len-1, 0);
 383
 384         if (format)
 385                 format_len = strlen(format) + 1;
 386         head = &marker_table[hash & ((1 << MARKER_HASH_BITS)-1)];
 387         hlist_for_each_entry(e, node, head, hlist) {
 388                 if (!strcmp(name, e->name)) {
 389                         printk(KERN_NOTICE
 390                                 "Marker %s busy\n", name);
 391                         return ERR_PTR(-EBUSY); /* Already there */
 392                 }
 393         }
 394         /*
 395          * Using kmalloc here to allocate a variable length element. Could
 396          * cause some memory fragmentation if overused.
 397          */
 398         e = kmalloc(sizeof(struct marker_entry) + name_len + format_len,
 399                         GFP_KERNEL);
 400         if (!e)
 401                 return ERR_PTR(-ENOMEM);
 402         memcpy(&e->name[0], name, name_len);
 403         if (format) {
 404                 e->format = &e->name[name_len];
 405                 memcpy(e->format, format, format_len);
 406                 if (strcmp(e->format, MARK_NOARGS) == 0)
 407                         e->call = marker_probe_cb_noarg;
 408                 else
 409                         e->call = marker_probe_cb;
 410                 trace_mark(core_marker_format, "name %s format %s",
 411                                 e->name, e->format);
 412         } else {
 413                 e->format = NULL;
 414                 e->call = marker_probe_cb;
 415         }
 416         e->single.func = __mark_empty_function;
 417         e->single.probe_private = NULL;
 418         e->multi = NULL;
 419         e->ptype = 0;
 420         e->format_allocated = 0;
 421         e->refcount = 0;
 422         e->rcu_pending = 0;
 423         hlist_add_head(&e->hlist, head);
 424         return e;
 425 }
 426
 427 /*
 428  * Remove the marker from the marker hash table. Must be called with mutex_lock
 429  * held.
 430  */
 431 static int remove_marker(const char *name)
 432 {
 433         struct hlist_head *head;
 434         struct hlist_node *node;
 435         struct marker_entry *e;
 436         int found = 0;
 437         size_t len = strlen(name) + 1;
 438         u32 hash = jhash(name, len-1, 0);
 439
 440         head = &marker_table[hash & ((1 << MARKER_HASH_BITS)-1)];
 441         hlist_for_each_entry(e, node, head, hlist) {
 442                 if (!strcmp(name, e->name)) {
 443                         found = 1;
 444                         break;
 445                 }
 446         }
 447         if (!found)
 448                 return -ENOENT;
 449         if (e->single.func != __mark_empty_function)
 450                 return -EBUSY;
 451         hlist_del(&e->hlist);
 452         if (e->format_allocated)
 453                 kfree(e->format);
 454         /* Make sure the call_rcu has been executed */
 455         if (e->rcu_pending)
 456                 rcu_barrier_sched();
 457         kfree(e);
 458         return 0;
 459 }
 460
 461 /*
 462  * Set the mark_entry format to the format found in the element.
 463  */
 464 static int marker_set_format(struct marker_entry *entry, const char *format)
 465 {
 466         entry->format = kstrdup(format, GFP_KERNEL);
 467         if (!entry->format)
 468                 return -ENOMEM;
 469         entry->format_allocated = 1;
 470
 471         trace_mark(core_marker_format, "name %s format %s",
 472                         entry->name, entry->format);
 473         return 0;
 474 }
 475
 476 /*
 477  * Sets the probe callback corresponding to one marker.
 478  */
 479 static int set_marker(struct marker_entry *entry, struct marker *elem,
 480                 int active)
 481 {
 482         int ret;
 483         WARN_ON(strcmp(entry->name, elem->name) != 0);
 484
 485         if (entry->format) {
 486                 if (strcmp(entry->format, elem->format) != 0) {
 487                         printk(KERN_NOTICE
 488                                 "Format mismatch for probe %s "
 489                                 "(%s), marker (%s)\n",
 490                                 entry->name,
 491                                 entry->format,
 492                                 elem->format);
 493                         return -EPERM;
 494                 }
 495         } else {
 496                 ret = marker_set_format(entry, elem->format);
 497                 if (ret)
 498                         return ret;
 499         }
 500
 501         /*
 502          * probe_cb setup (statically known) is done here. It is
 503          * asynchronous with the rest of execution, therefore we only
 504          * pass from a "safe" callback (with argument) to an "unsafe"
 505          * callback (does not set arguments).
 506          */
 507         elem->call = entry->call;
 508         /*
 509          * Sanity check :
 510          * We only update the single probe private data when the ptr is
 511          * set to a _non_ single probe! (0 -> 1 and N -> 1, N != 1)
 512          */
 513         WARN_ON(elem->single.func != __mark_empty_function
 514                 && elem->single.probe_private != entry->single.probe_private
 515                 && !elem->ptype);
 516         elem->single.probe_private = entry->single.probe_private;
 517         /*
 518          * Make sure the private data is valid when we update the
 519          * single probe ptr.
 520          */
 521         smp_wmb();
 522         elem->single.func = entry->single.func;
 523         /*
 524          * We also make sure that the new probe callbacks array is consistent
 525          * before setting a pointer to it.
 526          */
 527         rcu_assign_pointer(elem->multi, entry->multi);
 528         /*
 529          * Update the function or multi probe array pointer before setting the
 530          * ptype.
 531          */
 532         smp_wmb();
 533         elem->ptype = entry->ptype;
 534         elem->state = active;
 535
 536         return 0;
 537 }
 538
 539 /*
 540  * Disable a marker and its probe callback.
 541  * Note: only waiting an RCU period after setting elem->call to the empty
 542  * function insures that the original callback is not used anymore. This insured
 543  * by rcu_read_lock_sched around the call site.
 544  */
 545 static void disable_marker(struct marker *elem)
 546 {
 547         /* leave "call" as is. It is known statically. */
 548         elem->state = 0;
 549         elem->single.func = __mark_empty_function;
 550         /* Update the function before setting the ptype */
 551         smp_wmb();
 552         elem->ptype = 0;        /* single probe */
 553         /*
 554          * Leave the private data and id there, because removal is racy and
 555          * should be done only after an RCU period. These are never used until
 556          * the next initialization anyway.
 557          */
 558 }
 559
 560 /**
 561  * marker_update_probe_range - Update a probe range
 562  * @begin: beginning of the range
 563  * @end: end of the range
 564  *
 565  * Updates the probe callback corresponding to a range of markers.
 566  */
 567 void marker_update_probe_range(struct marker *begin,
 568         struct marker *end)
 569 {
 570         struct marker *iter;
 571         struct marker_entry *mark_entry;
 572
 573         mutex_lock(&markers_mutex);
 574         for (iter = begin; iter < end; iter++) {
 575                 mark_entry = get_marker(iter->name);
 576                 if (mark_entry) {
 577                         set_marker(mark_entry, iter, !!mark_entry->refcount);
 578                         /*
 579                          * ignore error, continue
 580                          */
 581                 } else {
 582                         disable_marker(iter);
 583                 }
 584         }
 585         mutex_unlock(&markers_mutex);
 586 }
 587
 588 /*
 589  * Update probes, removing the faulty probes.
 590  *
 591  * Internal callback only changed before the first probe is connected to it.
 592  * Single probe private data can only be changed on 0 -> 1 and 2 -> 1
 593  * transitions.  All other transitions will leave the old private data valid.
 594  * This makes the non-atomicity of the callback/private data updates valid.
 595  *
 596  * "special case" updates :
 597  * 0 -> 1 callback
 598  * 1 -> 0 callback
 599  * 1 -> 2 callbacks
 600  * 2 -> 1 callbacks
 601  * Other updates all behave the same, just like the 2 -> 3 or 3 -> 2 updates.
 602  * Site effect : marker_set_format may delete the marker entry (creating a
 603  * replacement).
 604  */
 605 static void marker_update_probes(void)
 606 {
 607         /* Core kernel markers */
 608         marker_update_probe_range(__start___markers, __stop___markers);
 609         /* Markers in modules. */
 610         module_update_markers();
 611 }
 612
 613 /**
 614  * marker_probe_register -  Connect a probe to a marker
 615  * @name: marker name
 616  * @format: format string
 617  * @probe: probe handler
 618  * @probe_private: probe private data
 619  *
 620  * private data must be a valid allocated memory address, or NULL.
 621  * Returns 0 if ok, error value on error.
 622  * The probe address must at least be aligned on the architecture pointer size.
 623  */
 624 int marker_probe_register(const char *name, const char *format,
 625                         marker_probe_func *probe, void *probe_private)
 626 {
 627         struct marker_entry *entry;
 628         int ret = 0;
 629         struct marker_probe_closure *old;
 630
 631         mutex_lock(&markers_mutex);
 632         entry = get_marker(name);
 633         if (!entry) {
 634                 entry = add_marker(name, format);
 635                 if (IS_ERR(entry))
 636                         ret = PTR_ERR(entry);
 637         } else if (format) {
 638                 if (!entry->format)
 639                         ret = marker_set_format(entry, format);
 640                 else if (strcmp(entry->format, format))
 641                         ret = -EPERM;
 642         }
 643         if (ret)
 644                 goto end;
 645
 646         /*
 647          * If we detect that a call_rcu is pending for this marker,
 648          * make sure it's executed now.
 649          */
 650         if (entry->rcu_pending)
 651                 rcu_barrier_sched();
 652         old = marker_entry_add_probe(entry, probe, probe_private);
 653         if (IS_ERR(old)) {
 654                 ret = PTR_ERR(old);
 655                 goto end;
 656         }
 657         mutex_unlock(&markers_mutex);
 658         marker_update_probes();
 659         mutex_lock(&markers_mutex);
 660         entry = get_marker(name);
 661         if (!entry)
 662                 goto end;
 663         if (entry->rcu_pending)
 664                 rcu_barrier_sched();
 665         entry->oldptr = old;
 666         entry->rcu_pending = 1;
 667         /* write rcu_pending before calling the RCU callback */
 668         smp_wmb();
 669         call_rcu_sched(&entry->rcu, free_old_closure);
 670 end:
 671         mutex_unlock(&markers_mutex);
 672         return ret;
 673 }
 674 EXPORT_SYMBOL_GPL(marker_probe_register);
 675
 676 /**
 677  * marker_probe_unregister -  Disconnect a probe from a marker
 678  * @name: marker name
 679  * @probe: probe function pointer
 680  * @probe_private: probe private data
 681  *
 682  * Returns the private data given to marker_probe_register, or an ERR_PTR().
 683  * We do not need to call a synchronize_sched to make sure the probes have
 684  * finished running before doing a module unload, because the module unload
 685  * itself uses stop_machine(), which insures that every preempt disabled section
 686  * have finished.
 687  */
 688 int marker_probe_unregister(const char *name,
 689         marker_probe_func *probe, void *probe_private)
 690 {
 691         struct marker_entry *entry;
 692         struct marker_probe_closure *old;
 693         int ret = -ENOENT;
 694
 695         mutex_lock(&markers_mutex);
 696         entry = get_marker(name);
 697         if (!entry)
 698                 goto end;
 699         if (entry->rcu_pending)
 700                 rcu_barrier_sched();
 701         old = marker_entry_remove_probe(entry, probe, probe_private);
 702         mutex_unlock(&markers_mutex);
 703         marker_update_probes();
 704         mutex_lock(&markers_mutex);
 705         entry = get_marker(name);
 706         if (!entry)
 707                 goto end;
 708         if (entry->rcu_pending)
 709                 rcu_barrier_sched();
 710         entry->oldptr = old;
 711         entry->rcu_pending = 1;
 712         /* write rcu_pending before calling the RCU callback */
 713         smp_wmb();
 714         call_rcu_sched(&entry->rcu, free_old_closure);
 715         remove_marker(name);    /* Ignore busy error message */
 716         ret = 0;
 717 end:
 718         mutex_unlock(&markers_mutex);
 719         return ret;
 720 }
 721 EXPORT_SYMBOL_GPL(marker_probe_unregister);
 722
 723 static struct marker_entry *
 724 get_marker_from_private_data(marker_probe_func *probe, void *probe_private)
 725 {
 726         struct marker_entry *entry;
 727         unsigned int i;
 728         struct hlist_head *head;
 729         struct hlist_node *node;
 730
 731         for (i = 0; i < MARKER_TABLE_SIZE; i++) {
 732                 head = &marker_table[i];
 733                 hlist_for_each_entry(entry, node, head, hlist) {
 734                         if (!entry->ptype) {
 735                                 if (entry->single.func == probe
 736                                                 && entry->single.probe_private
 737                                                 == probe_private)
 738                                         return entry;
 739                         } else {
 740                                 struct marker_probe_closure *closure;
 741                                 closure = entry->multi;
 742                                 for (i = 0; closure[i].func; i++) {
 743                                         if (closure[i].func == probe &&
 744                                                         closure[i].probe_private
 745                                                         == probe_private)
 746                                                 return entry;
 747                                 }
 748                         }
 749                 }
 750         }
 751         return NULL;
 752 }
 753
 754 /**
 755  * marker_probe_unregister_private_data -  Disconnect a probe from a marker
 756  * @probe: probe function
 757  * @probe_private: probe private data
 758  *
 759  * Unregister a probe by providing the registered private data.
 760  * Only removes the first marker found in hash table.
 761  * Return 0 on success or error value.
 762  * We do not need to call a synchronize_sched to make sure the probes have
 763  * finished running before doing a module unload, because the module unload
 764  * itself uses stop_machine(), which insures that every preempt disabled section
 765  * have finished.
 766  */
 767 int marker_probe_unregister_private_data(marker_probe_func *probe,
 768                 void *probe_private)
 769 {
 770         struct marker_entry *entry;
 771         int ret = 0;
 772         struct marker_probe_closure *old;
 773
 774         mutex_lock(&markers_mutex);
 775         entry = get_marker_from_private_data(probe, probe_private);
 776         if (!entry) {
 777                 ret = -ENOENT;
 778                 goto end;
 779         }
 780         if (entry->rcu_pending)
 781                 rcu_barrier_sched();
 782         old = marker_entry_remove_probe(entry, NULL, probe_private);
 783         mutex_unlock(&markers_mutex);
 784         marker_update_probes();
 785         mutex_lock(&markers_mutex);
 786         entry = get_marker_from_private_data(probe, probe_private);
 787         if (!entry)
 788                 goto end;
 789         if (entry->rcu_pending)
 790                 rcu_barrier_sched();
 791         entry->oldptr = old;
 792         entry->rcu_pending = 1;
 793         /* write rcu_pending before calling the RCU callback */
 794         smp_wmb();
 795         call_rcu_sched(&entry->rcu, free_old_closure);
 796         remove_marker(entry->name);     /* Ignore busy error message */
 797 end:
 798         mutex_unlock(&markers_mutex);
 799         return ret;
 800 }
 801 EXPORT_SYMBOL_GPL(marker_probe_unregister_private_data);
 802
 803 /**
 804  * marker_get_private_data - Get a marker's probe private data
 805  * @name: marker name
 806  * @probe: probe to match
 807  * @num: get the nth matching probe's private data
 808  *
 809  * Returns the nth private data pointer (starting from 0) matching, or an
 810  * ERR_PTR.
 811  * Returns the private data pointer, or an ERR_PTR.
 812  * The private data pointer should _only_ be dereferenced if the caller is the
 813  * owner of the data, or its content could vanish. This is mostly used to
 814  * confirm that a caller is the owner of a registered probe.
 815  */
 816 void *marker_get_private_data(const char *name, marker_probe_func *probe,
 817                 int num)
 818 {
 819         struct hlist_head *head;
 820         struct hlist_node *node;
 821         struct marker_entry *e;
 822         size_t name_len = strlen(name) + 1;
 823         u32 hash = jhash(name, name_len-1, 0);
 824         int i;
 825
 826         head = &marker_table[hash & ((1 << MARKER_HASH_BITS)-1)];
 827         hlist_for_each_entry(e, node, head, hlist) {
 828                 if (!strcmp(name, e->name)) {
 829                         if (!e->ptype) {
 830                                 if (num == 0 && e->single.func == probe)
 831                                         return e->single.probe_private;
 832                         } else {
 833                                 struct marker_probe_closure *closure;
 834                                 int match = 0;
 835                                 closure = e->multi;
 836                                 for (i = 0; closure[i].func; i++) {
 837                                         if (closure[i].func != probe)
 838                                                 continue;
 839                                         if (match++ == num)
 840                                                 return closure[i].probe_private;
 841                                 }
 842                         }
 843                         break;
 844                 }
 845         }
 846         return ERR_PTR(-ENOENT);
 847 }
 848 EXPORT_SYMBOL_GPL(marker_get_private_data);